Key Takeaways
- Viggle JST-2 is a physics-native 3D world foundation model that generates 3D scenes, characters, and motions from images, text, or video in seconds.
- It is built entirely in 3D space rather than pixel or video space, which lets it treat scenes, characters, and motion as one unified system.
- You can try it for free via the official Try JST-2 page, which leads to the PINOC browser interface with Scene and Agent modes.
- Outputs include Gaussian Splat scenes and GLB or FBX motion data, ready for downstream 3D tools and game engines.
Quick Facts
| Type | Physics-native 3D world foundation model |
|---|---|
| Devices | Web browser (PINOC interface) |
| Price | Free to try (official claim) |
| Languages | Check official site |
| Highlight | Generates 3D scenes, characters, and motion in seconds |
| Learning curve | Moderate; basic 3D familiarity helps |
| Category | AI & Models |
| Creator | ViggleAI |
| Official site | viggle.ai ↗ |
| Heat | 🔥 1.7M Heat |
| Last updated |
What Makes JST-2 a Physics-Native 3D World Model
ViggleAI released JST-2 on October 19, 2026, describing it as the world's first full-stack physics-native world foundation model. Most world models learn in pixel or video space, predicting frames or patches. JST-2 instead builds its entire model stack in 3D, starting from first principles. That means it reasons about geometry, scale, and physical plausibility directly, rather than trying to infer 3D structure from 2D images. The result is a system that can generate a 3D scene in seconds and expand it in real time.
The core idea is unification. Instead of separate pipelines for environments, characters, and animation, JST-2 uses one general architecture that handles all three. You can create a character with motion already built in, or generate a full-body 3D motion from text or video. Because everything lives in the same 3D space, scenes and characters compose naturally. This is a departure from tools that stitch together 2D assets or rely on video diffusion, and it is why ViggleAI calls it a world foundation model rather than just a 3D generator.
- 3D-first architecture
The model operates in 3D space from the ground up, which helps it maintain consistent geometry and physics across generated content.
- Unified scene, character, and motion
One model handles environments, full-body characters, and animation, so you do not need to switch between separate tools.
- Fast generation
ViggleAI claims second-level scene generation and real-time expansion, making it suitable for rapid iteration.
How JST-2 Works: Scene Mode, Agent Mode, and Outputs
You access JST-2 through the official Try JST-2 page, which opens the PINOC browser interface. PINOC has two main modes. Scene mode lets you build environments, starting from an image, text prompt, or video. Agent mode is where you combine scenes, characters, and motions into a coherent world. The agent likely helps orchestrate the generation process, though the exact controls are best explored directly in the interface.
JST-2 outputs Gaussian Splat scenes, a format that represents 3D geometry and appearance with high visual fidelity, and motion data in GLB or FBX formats. Those formats are widely supported by 3D tools and game engines, so you can take generated assets into Blender, Unity, Unreal Engine, or similar software for further editing. The model can generate full-body 3D motion from text or video, meaning you can describe an action or provide a reference clip and get a 3D animation back.
- Scene mode
Build indoor or outdoor environments from images, text, or video. Start small to gauge quality and wait times.
- Agent mode
Combine scenes, characters, and motions. This is where you assemble a complete world with multiple elements.
- Export formats
Gaussian Splat scenes and GLB or FBX motion data are ready for downstream 3D pipelines.
Getting Started with JST-2: A Practical Workflow
Begin with a simple test. Open the Try JST-2 page and use Scene mode to generate a small indoor room from a single image or a short text prompt. This lets you judge generation quality, wait times, and how well the model handles geometry before you invest time in a complex project. Once you are comfortable, move to a single character and generate a motion from text, such as a walk cycle or a wave. Export the motion as GLB or FBX and inspect it in your 3D tool of choice.
When you are ready for more, switch to Agent mode and combine your scene and character. Add motion and see how the pieces integrate. Because JST-2 treats everything as one 3D system, you can iterate quickly: adjust the scene, swap motions, or add characters without rebuilding from scratch. Keep an eye on the official site for updates, since ViggleAI may refine the interface and add features over time. The free trial is a good way to evaluate whether JST-2 fits your workflow before committing to larger projects.
- Start small
Test with a single room or character to understand quality and speed before scaling up.
- Use reference media
Images and video can guide generation; a clear reference often improves results.
- Export and refine
Take GLB or FBX motion data into your 3D software for polishing and integration.
JST-2 vs. Other World Models and Who It Is For
Most world models, such as those focused on video prediction, learn in pixel or video space. They generate frames that look realistic but may lack consistent 3D structure. JST-2's 3D-native approach gives it an edge in tasks that require spatial coherence, like building game levels or animating characters that must move believably through an environment. However, it is not a replacement for a full game engine or a traditional 3D modeling suite; it is a generator that feeds those pipelines.
JST-2 is aimed at game developers, prototypers, and 3D artists who need rapid ideation. It is also useful for previsualization and concept art. Limitations include the lack of a dedicated product page and the fact that some details, such as exact pricing beyond the free trial and supported languages, are not yet confirmed. Check the official site at viggle.ai for the latest information. If you need production-ready assets with fine control, you will still rely on traditional tools, but JST-2 can dramatically speed up the early stages.
- Compared to video-based world models
JST-2 generates true 3D structure, which is better for spatial consistency and downstream 3D work.
- Best for prototyping
Use it to quickly block out scenes, characters, and motions before committing to detailed modeling.
- Not a full engine
JST-2 outputs assets; you still need a game engine or 3D software to build the final experience.
Step-by-Step Guide
Visit the official Try JST-2 page
Go to viggle.ai and look for the Try JST-2 link. It opens the PINOC browser interface, where you can start generating. No dedicated product page exists yet, so the official site is your entry point.
Generate a simple scene in Scene mode
Upload an image or type a text prompt to create a small indoor room. This first test helps you understand generation quality and wait times. Keep the scene simple to avoid long processing.
Create a character with motion
Use an image to generate a full-body 3D character, or provide text or video to create a motion. Export the motion as GLB or FBX. Inspect the result in your 3D software to check rigging and animation quality.
Combine elements in Agent mode
Switch to Agent mode to bring your scene and character together. Add motions and adjust placement. This is where JST-2's unified 3D system shines, letting you iterate quickly without rebuilding.
Export and integrate into your pipeline
Download your Gaussian Splat scene and motion files. Import them into Blender, Unity, Unreal Engine, or another 3D tool. Use them as a base for further modeling, texturing, and animation.
Frequently Asked Questions
Is Viggle JST-2 free to use?
ViggleAI states that JST-2 is free to try, but the exact limits of the free tier are not specified. There may be usage caps or watermarks, and paid plans could exist. Check the official site at viggle.ai for current pricing and terms before relying on it for a project.
What devices can I use to access JST-2?
JST-2 runs in a web browser through the PINOC interface, so it works on desktop and laptop computers. A stable internet connection is required. Mobile support is not confirmed; for serious 3D work, a desktop browser is recommended for better performance and control.
How does JST-2 differ from other AI video generators?
Most AI video generators work in pixel or video space, creating 2D frames that may lack consistent 3D structure. JST-2 is built entirely in 3D space, so it generates true 3D scenes, characters, and motions. This makes its outputs directly usable in 3D pipelines and game engines, rather than requiring conversion from video.
What file formats does JST-2 export?
JST-2 exports Gaussian Splat scenes for environments and GLB or FBX files for motion data. These formats are widely supported by 3D software and game engines, including Blender, Unity, and Unreal Engine. You can import them for further editing and integration.
Can I use JST-2 for commercial projects?
Commercial usage rights depend on ViggleAI's terms of service, which are not detailed in the announcement. You should review the official terms on viggle.ai before using generated assets in a commercial product. When in doubt, contact the company directly for clarification.
