World building guide
How JST-2 expands a 3D world
JST-2 world expansion uses existing views as context for generating more of a scene. Viggle also describes reconstructing environments from sparse real observations. Both workflows need more than a fresh text prompt.
New views extend the spatial context
Viggle describes a spatial context window of up to 64 input views. As a world expands one view at a time, the new view contributes context for subsequent generation. The intent is to build on a shared environment rather than generate unrelated pictures of similar places.
This context limit is a model capability, not a promise of unlimited coherent terrain. When assessing an expanded scene, revisit recognizable landmarks and check that doors, walls and object positions remain plausible across the transition.
Moving the camera is different from adding content
JST-2 decodes its representation into Gaussian splats for local browser rendering. You can inspect an already generated scene from different viewpoints without rerunning the model. New content requires inference; camera movement alone does not necessarily expand the world.
For a practical review, first explore the existing result. Identify where coverage ends, then inspect the output of any expansion operation separately. This helps distinguish a rendering artifact in existing content from a newly generated region that does not match its surroundings.
Reconstruct a real place from limited observations
Viggle’s release demonstrates reconstructing complete environments from sparse real-world observations. The inputs provide evidence about appearance and structure, while the model supplies an interpretation of portions that were not directly observed. A convincing novel view is not a measurement of unseen geometry.
For evaluation, keep reference views that were not used as input and compare the reconstruction against them. Pay attention to door placement, repeated structures and reflective surfaces. Mapping, construction and simulation applications require their own accuracy checks before relying on generated content.
Separate expansion today from text guidance tomorrow
The official developer page describes generating new camera views and rebuilding a scene from image batches. Consult its current documentation for access and supported inputs. The release separately labels text prompts that guide what appears during expansion as an upcoming feature.
A version with more splats per scene is also described as upcoming. Neither statement guarantees a release date. Scene collisions and object interactions remain rule-based in JST-2; Viggle’s learned-interaction physics belongs to its future JST-3 research direction.
Questions & answers
Can I expand a scene in JST-2 Studio?
This studio’s text-scene form creates a scene from a prompt and exposes its returned assets. It does not offer a world-expansion or reconstruction control. Use Viggle’s official PINOC or developer resources to assess those workflows.
Does 64 input views mean I must supply 64 images?
No. Viggle describes up to 64 views as a spatial context window. It is not a requirement to upload that many images, and it does not specify the image count accepted by every product interface.
Can I tell expansion to add a particular room?
The JST-2 launch article identifies text-guided expansion as upcoming. Do not assume a prompt can direct an expansion in every currently available interface. Check the official product and documentation for its current controls.
Check the original sources.
Reviewed . Model claims are attributed to Viggle; practical suggestions are editorial guidance.