Source library
Choose backgrounds and objects independently.
All and Half apply to one source group. Half selects alternating independent sources in a stable semantic order. Each task uses its eligible objects.
Sources, preparation & task suitability
Photographic and GPT-generated counterparts share a semantic category; composition, lighting and geometry differ. Placement, movement and controlled signals are programmatic. Source dimensions are separate from stimulus and model input dimensions.
| Asset | Origin | Source / prepared pixels | Eligible tasks | Files and attribution |
|---|