Viggle says its JST-2 model builds a 3D world in about a second
The company behind the meme-making app says JST-2 makes scenes, characters and motion you can walk around in. The speed and comparison numbers come from Viggle.
On October 19, 2026, Viggle introduced JST-2. Viggle is the startup behind a meme-making app it says has passed 50 million users. It calls JST-2 a "physics-native world foundation model." In plain terms, it is meant to build 3D scenes, characters and movement that share one world you can explore.
How Viggle says it works
Many AI video and world models learn from flat pixels. Viggle says JST, short for Joint Space-and-Time, models the world as lasting 3D "atoms" instead. Each one stands for a bit of matter or part of an object, with a position in space and learned traits for how it looks and moves. Viggle says that lets objects stay put when you look away and keep their identity as they move.
The model can start from a prompt or from photos of a real place. It outputs images or 3D Gaussian splats, a format that renders in your browser. Viggle says once a scene is made, you can look around it again and again without rerunning the model.
The claims
Viggle says JST-2 makes a 3D scene in about one second, for less than a cent, timed on NVIDIA B200 chips. It compares that with about five to six minutes for a rival model called Marble. That is Viggle's own timing.
Viggle also says JST-2 can turn a single image into a 3D character that is ready to animate without the usual rigging step, and can turn text or video into full-body motion.
In Viggle's own blind tests, human judges preferred JST-2's characters over those from Meshy 7.1 and Tripo H3.1, especially on faces. Viggle's tests also had JST-2 ahead of Cartwheel, HY-Motion-1.0 and DeepMotion on motion. Those are Viggle's tests and its scoring, which counts a draw as half a win.
Shipped, and still coming
JST-2 is the shipped part. Viggle says you can try it free in the browser through its PINOC app, in a Scene mode or an Agent mode that combines scenes, characters and motion.
The rest is a plan. Viggle says later releases will let text prompts guide how a scene grows and will add finer detail. It says it is working on JST-3, which would replace today's rule-based physics for collisions and object interactions with learned physics. No date is given.
Source: Viggle