01 · Primary object记录
Media asset
An image, video, audio item, or screenshot reference with a type and explicit file-delivery state.
- asset_id & type
- asset_url or file_ref
- format & dimensions
Images, video & audio data
Media data with the source context still attached.
Collect documented image and video search metadata, reverse-image results, Maps photo references, and browser screenshots—or scope a maintained multimodal dataset that keeps files, metadata, renditions, provenance, and derived labels distinct.
Multimodal object model
A media result is more than a URL.
Separate the file or media binary from its observed metadata, source page, renditions, text, rights context, and derived labels. Each object can then change without corrupting the others.
02 · Discovery context记录
Source page or result
The public page or search result where the asset and its metadata were observed.
- source_url
- query & position
- publisher context
03 · Related objects飞行员先
Rendition, text & label
Alternate renditions, captions, transcripts, alt text, and derived labels remain linked records with their own provenance.
- rendition relation
- caption or transcript
- label method
Interpretation boundaryNot assumed
Metadata is not media rights.
A discovered URL, caption, thumbnail, or delivered file does not establish ownership or permission to reuse it. Rights, licensing, attribution, and downstream use remain the customer's responsibility.
Coverage brief
Start from a documented surface, then confirm the asset contract.
Coverage describes the source, result or page family, request context, metadata fields, and whether file acquisition is included. It is never a blanket claim about media across the web.
Engine or family输入环境国家
Asset manifest
Make file state and provenance inspectable.
The schema tells consumers whether a record describes a media reference or a delivered file, where it came from, which rendition was observed, and which enrichment is derived.
Identity & file state
- asset_id
- Delivery identity for the media object.
- media_type
- Image, video, audio, or screenshot.
- file_state
- Referenced, delivered, unavailable, or out of scope.
Observed attributes
- asset_url
- Source-exposed media or thumbnail reference.
- format
- Observed or detected media format.
- dimensions
- Width, height, duration, or another applicable property.
Source, text & renditions
- source_page
- Public page or result containing the observation.
- rendition_of
- Parent relationship when the contract establishes one.
- text_ref
- Caption, transcript, alt text, or label record.
Traceability & responsibility
- observed_at
- Time associated with the source observation.
- derivation_method
- Source, parser, model, or rule that produced a field.
- rights_state
- Customer-review state, never a license assertion.
{
"asset_id": "media_001",
"media_type": "image",
"file": {
"state": "not_requested",
"storage_ref": null
},
"metadata": {
"asset_url": "https://media.example/image.jpg",
"width": 1200,
"height": 800,
"caption": "Example caption"
},
"source": {
"page_url": "https://publisher.example/page",
"surface": "image_search"
},
"rights_state": "customer_review",
"observed_at": "YYYY-MM-DDThh:mm:ssZ",
"schema_version": "media.v1"
}Identity & relationships
Link renditions without pretending they are identical.
Keep every source reference first. Exact file relationships and perceptual candidates are different claims and must remain distinguishable.
- 01
Source key
URL + source context
Retained as the observed reference. - 02
Exact evidence
File hash or source ID
Used only when the file or key is available. - 03
Candidate evidence
Dimensions + perceptual cues
Pilot-first rules flag likely renditions. - 04
Relationship state
Exact · candidate · distinct
Ambiguity remains visible for review.
Freshness & history
Timestamp discovery, file access, and enrichment separately.
A source result, referenced media file, transcript, and derived label may be produced at different times. One generic “updated” field cannot describe them honestly.
- Requestedrequested_at
The source, input, locale, and requested asset scope entered collection.
- 观察到observed_at
The page or result exposed the metadata or reference.
- Acquiredfile_acquired_at
A media binary was delivered when file acquisition is in scope.
- Changedfirst_seen / last_seen
The source reference, rendition, or metadata entered tracked history.
Quality & missingness
Do not turn “file not delivered” into “media absent.”
Quality states distinguish source absence, reference-only scope, a failed acquisition, unsupported media, and an enrichment that has not run.
Manifest passes contract
Required metadata, provenance, type, and state fields are present.
Metadata without a file
The page exposed a media reference; binary delivery was not requested.
Rendition match is ambiguous
A candidate relationship remains inspectable rather than silently merged.
No supported observation
Source absence, access failure, and out-of-scope media retain different reasons.
Structural checks
Required keys, types, URL shape, file-state consistency, and schema version.
Media checks
Format, dimensions, duration, hash, decode state, or transcript presence only where contracted.
Customer acceptance
The customer approves source eligibility, rights workflow, label taxonomy, usefulness thresholds, and downstream use.
Operating model
Choose who operates each part of the media-data lifecycle.
Compare access, extraction, file handling, enrichment, monitoring, and delivery before choosing infrastructure or an operated dataset.
Maximum control
Build collection and media handling on proxy infrastructure.
Your team chooses sources, retrieves pages or files, extracts metadata, manages storage, labels, deduplication, rights review, quality, and delivery.
责任拥有者
Source, object, and rights briefCustomer
Network access infrastructureWSA
Extraction and file handlingCustomer
Labels, quality, and maintenanceCustomer
Storage, rights review, and useCustomer
Applications
Build workflows on inspectable media objects.
Each application uses the same asset, source, rendition, time, and provenance foundation while the customer owns interpretation and permitted use.
Visual search research
Observe image and video result metadata for defined queries and markets.
query · result · asset reference · source · positionBrand asset discovery
Find public media references for review without treating discovery as a rights conclusion.
asset · source page · caption · observed_atCatalog media QA
Compare expected media references, formats, and dimensions across public product surfaces.
entity cue · asset · rendition · quality stateContent monitoring
Track first-seen and last-seen observations for agreed public media surfaces.
source · asset · observation · historyMultimodal AI inputs
Scope linked files, metadata, text, and labels against an approved public-source brief.
file · metadata · transcript · label provenanceScreenshot evidence
Capture rendered public pages with request context and observation time attached.
URL · viewport · screenshot · timestamp · stateRepresentative pilot
Prove the manifest before scaling the media volume.
Sample ordinary and difficult assets, clarify file-versus-metadata delivery, test provenance and missingness, and confirm rights responsibilities before a recurring program.
- 01
Define
Name public sources, media types, files, metadata, transformations, cadence, and exclusions.
- 02
Sample
Include missing files, alternate renditions, duplicate candidates, unsupported formats, and changed pages.
- 03
Inspect
Review manifests, files where included, labels, provenance, timestamps, and collection states.
- 04
Accept
Agree schema, identity rules, quality checks, rights boundary, delivery, and maintenance response.
Evaluation FAQ
Resolve the media contract before collection begins.
These answers distinguish documented access from pilot-first enrichment and keep file delivery separate from rights and interpretation.
What image, video, and audio data can WebScrapingAPI provide?
Documented products can return metadata from Google image and video result surfaces, reverse-image results, Google Maps photo references, and browser screenshots. Broader media-file acquisition, audio extraction, transcripts, labels, deduplication, and training-ready datasets begin with a representative pilot.
Do records include the media file or only metadata?
The contract states this explicitly. A record can describe a source page, media URL, thumbnail, dimensions, format, caption, or rendition without delivering the underlying media binary. When file delivery is in scope, the file and its metadata remain separate, linked objects with their own states.
Does delivered metadata include permission to reuse the media?
No. Metadata is not media rights. Copyright, licensing, attribution, permitted use, retention, and downstream distribution remain the customer's responsibility; WebScrapingAPI does not provide a copyright-clear guarantee.
How are images, videos, audio, and renditions related?
The model keeps the media asset separate from the page or result where it appeared, its observed metadata, alternate renditions, captions or transcripts, source and rights context, and any derived label. Relationships and provenance are retained rather than flattened into one ambiguous row.
Can WebScrapingAPI collect private media?
Private accounts, restricted libraries, direct messages, and other non-public media are not standard scope. Collection is limited to agreed eligible public web surfaces and the access context confirmed during scoping.
Can you create labeled multimodal training datasets?
Labeling, media deduplication, transcript enrichment, and training-dataset assembly are pilot-first capabilities. A pilot defines the public sources, label taxonomy, file and metadata boundary, rights responsibilities, acceptance sample, and delivery format before production scope is agreed.
How is duplicate media handled?
Source URLs and exact file hashes can support deterministic relationships when files are delivered. Perceptual matching across resized, cropped, re-encoded, or excerpted media is a pilot-first rule set, and ambiguous candidates stay visible instead of being silently merged.
What do missing captions, transcripts, or files mean?
The record distinguishes a value not exposed by the source, a rendition unavailable in the requested context, a file outside the contracted scope, a collection failure, and a field awaiting review. Missing values are not automatically treated as empty content.
How fresh can multimodal observations be?
Request-time documented APIs return an observation for the submitted context. Scheduled and managed programs use a source-specific cadence agreed after sampling; observed_at, first_seen, last_seen, and delivery timestamps remain distinct.
Who maintains extraction and quality monitoring?
Proxy customers own their collectors, parsers, and monitoring. WebScrapingAPI maintains documented API behavior within the product boundary. In scheduled or managed delivery, WebScrapingAPI can own the contracted extraction, schema, quality checks, source-change maintenance, and delivery operation.
Images, video & audio data
Move from media record design to an operated video-data supply.
Use this guide to define the object your system needs, then continue to Video Data for AI for scenario-focused clips, audio, transcripts, metadata, and recurring delivery.