A FOCUSED PROTOTYPE
TwinFrame
Arrange several photos into a navigable, manually composed spatial scene.
Open Demo01 / Problem
A set of photos can suggest how a place feels, but a flat list makes their relationship difficult to explore.
02 / Design
Give each photo a place in an explicitly conceptual scene. Make navigation useful without pretending the photo pixels reveal actual geometry.
03 / Build
Local images become bounded GPU textures on framed panels in a Three.js scene. A perspective camera, lighting, depth testing, direct orbit and selection support exploration. Textures and geometry are released on removal, reset and navigation; a labeled 2D fallback keeps the tool usable without WebGL.
04 / Demonstration
Open the sample scene or add several photos, adjust their order, move the viewpoint, and export a record of the scene.
05 / Limitations
An approximation, not photogrammetry, a precise scan or a recovered 3D model. Photos are manually arranged.
06 / Next step
Evaluate whether manual photo placement helps people communicate spaces before considering calibrated reconstruction research.
Exact technology
HTML, CSS, JavaScript ES modules, Three.js 0.180.0 (MIT), WebGL 2, Canvas 2D, File and Blob APIs. Same-origin, pinned Three.js with its MIT license; all images stay local.
Evidence boundary
This case study describes prototype behavior and design decisions. It does not claim production deployment by customers, measured commercial outcomes or scientific validation.