Manifest Specification (scrollsequence-manifest.json)
The Asset Pipeline generates a standard JSON manifest connecting your processed images and depth data directly to the rendering engine.
Specification Example
{
"version": "2.0.0",
"input": {
"path": "input.mp4",
"width": 1920,
"height": 1080,
"frameCount": 240
},
"imageSizes": {
"1280": {
"width": 1280,
"height": 720,
"files": {
"i": "1280/%04d.jpg",
"d": "depthmaps/%04d.jpg"
},
"frameCount": 240
},
"1920": {
"width": 1920,
"height": 1080,
"files": {
"i": "1920/%04d.jpg",
"d": "depthmaps/%04d.jpg"
},
"frameCount": 240
}
},
"tracking": {
"main": [
{ "frame": 0, "x": 0.5, "y": 0.5, "scale": 0.1 }
]
},
"vision": {
"camera": { "movement": "zoom-in", "shot_type": "medium" },
"subjects": [
{ "subject_id": "car", "description": "sports car" }
]
}
}Manifest Fields Reference
input
Metadata describing the original raw video source prior to conversion:
path: Original source filename.width/height: Source video resolution in pixels.frameCount: Total number of frames extracted.
imageSizes
Keyed dictionary containing each generated resolution tier (e.g., "1280", "1920"):
width/height: Output resolution dimensions for this tier.frameCount: Number of frames present in this tier.files.i: Zero-padded printf pattern (e.g.1280/%04d.jpgformats to1280/0000.jpg,1280/0001.jpg, etc.).files.d: Relative path pattern pointing to corresponding 3D grayscale depth maps (e.g.depthmaps/%04d.jpg).
tracking (Optional, Cloud AI)
Bounding box and center-point coordinates for subject tracking across frames. Allows anchoring dynamic HTML callouts directly to moving objects.
vision (Optional, Cloud AI)
Structured video understanding data generated by AI vision analysis:
camera: Inferred camera motion (pan, tilt, zoom, stationary).subjects: List of detected primary subjects and semantic descriptions.
Last updated on