City and witnesses
Tests translucent architecture, a crowd at several scales, reflective water, and ambiguous foreground figures.
- Source SHA-256
- 7d2f720654c3…d544bd6c
- Scene
- not generated
Spatial media experiment · source preserving
This page is deliberately not a gallery of flat images made to look like point clouds. A result belongs here only when it is a real renderable spatial artifact, preserves the source beside it, and discloses what the single-image reconstruction had to invent.
The sources were selected to expose different failure modes in a single-image spatial interpretation: architectural depth, human scale, occlusion, reflective water, nested geometry, and large regions the source never shows.
Tests translucent architecture, a crowd at several scales, reflective water, and ambiguous foreground figures.
Tests a clear foreground subject, an implied path, a distant destination, and geometry that must occlude correctly as the camera moves.
Tests nested arcs, negative space, a small human scale reference, and whether depth survives without color doing the separation work.
The acceptance boundary is intentionally stricter than visual resemblance.
The result must include a loadable spatial file, planned as .spz. A PNG, video, or still render alone is not the scene.
Translation must produce meaningful parallax, changing occlusion, and a different spatial relationship among visible elements.
The original artwork stays canonical and is shown beside the generated scene. The experiment does not overwrite or silently “upgrade” it.
The record must identify the generator commit, provider and model versions, parameters, output hash, and inferred backsides, holes, floating fragments, or duplicated forms.
Source selection and receipts are complete. Scene generation is not.
neilsonnn/image-blaster requires the upstream service credentials to be configured locally or in a protected secret store. A single-image result will be described as a generated spatial interpretation, not a scan or measured reconstruction.