Video Data Extraction API
Define a schema. Get structured JSON. Extract entities, insights, and facts from any video or uploaded file. Supports YouTube, Instagram, TikTok, Facebook, X, Rumble, Vimeo, Dailymotion, and Snapchat — powered by a 2-phase AI pipeline with intelligent prompt caching.
- What is the Video Data Extraction API?
- The Video Data Extraction API is a schema-driven endpoint that turns any video into Pydantic-validated structured JSON. You define the fields you need; VidNavigator transcribes, analyzes, and extracts them through a 2-phase AI pipeline with a 2-hour shared prompt cache — replacing brittle LLM prompt engineering with guaranteed schema conformance.
Why Extract API?
- Custom schema extraction — define exactly which fields you want back (String, Number, Boolean, Array, Object, Enum)
- Built-in auto-transcription — videos without platform captions (Instagram, TikTok, Facebook, X, etc.) are automatically transcribed via speech-to-text, no separate API call needed
- Video metadata included — every response includes video info (title, channel, duration, views, publish date) alongside your extracted data
- Prompt caching — your compiled extraction prompt is cached for 2 hours and reused across videos, making repeat extractions faster and cheaper
- Broad platform support — extract from online videos on YouTube, Instagram, TikTok, Facebook, X, Rumble, Vimeo, Dailymotion, Snapchat, and Loom, or use
/v1/extract/filefor uploaded audio/video files - Multilingual output — results are returned in the same language as your schema descriptions and field definitions, supporting 99+ languages
How It Works — 2-Phase Pipeline
Prompt Compilation
Cached- You send a schema + optional
what_to_extractinstruction - An AI prompt engineer generates an optimized system prompt and user prompt template
- The compiled plan is cached with a 2-hour TTL based on a fingerprint of your schema + instructions — within that window, identical schemas reuse the same plan instantly
Structured Extraction
Per video- The cached prompt is hydrated with the video transcript
- AI extracts data following your schema strictly (Pydantic-enforced output)
- You receive clean, validated JSON matching your exact schema
Use Case Templates
Define your extraction schema in JSON or YAML. Here are ready-to-use templates for common use cases.
Lead Generation
Market Research
Content & Creator Analysis
AI Pipeline / RAG Ingestion
Brand & E-Commerce Monitoring
Fact-Checking & Claim Extraction
Quickstart
Call the Extract API with a video URL and your schema. The response contains structured JSON matching your exact field definitions. Each call processes one video at a time — to extract from multiple videos, make one API call per video.
Extract from Uploaded Files
The /v1/extract/file endpoint works exactly like /v1/extract/video but takes a file_id instead of a video_url. Files must be uploaded and transcribed first using the transcribe endpoint.
Built-In Auto-Transcription
For non-YouTube platforms, the Extract API automatically transcribes the video audio when no platform transcript exists. This is enabled by default via the transcribe parameter — no separate API call needed.
- •Speech-to-text credits are charged based on the video's duration — same rate as
/v1/transcribe. - •Transcripts are cached — subsequent extractions on the same video reuse it at no extra cost.
- •If either transcription or extraction fails, all charges are reverted automatically.
- •Set
transcribe=falseto disable auto-transcription and require an existing transcript. - •YouTube videos rely on platform captions — auto-transcription is not available for YouTube.
Schema Rules
- Max 10 root fields
- Max 3 nesting levels (level 3 must be primitive)
- Max 10 subfields per Object
- Every field needs type and description
- Supported types: String, Number, Boolean, Integer, Array, Object, Enum
Example Response
The API returns clean, validated JSON that matches your schema exactly.
Prompt Caching Deep Dive
The compiled prompt (system prompt + user prompt template) is cached with a 2-hour TTL based on a fingerprint of your schema and instructions.
~2-3 s overhead to generate an optimized extraction prompt from your schema. Subsequent calls within the 2-hour cache window skip compilation entirely.
Within the 2-hour TTL window, every call with the same schema and instructions reuses the cached prompt instantly.
Changing the schema or instructions creates a fresh cached plan automatically.
Define once, extract from thousands of videos within the 2-hour cache window. Cached plans are not tied to a single video.
Pricing
Each extraction counts as 1 video analysis for standard-length videos. For longer transcripts, billing scales as ceil(total_tokens / 15,000) analysis credits. If auto-transcription is triggered, speech-to-text hours are also charged based on video duration. All charges are reverted if the request fails.
Frequently Asked Questions
What video platforms are supported?
What happens if the video has no transcript?
transcribe=false.Is the extraction prompt really cached?
Can I use YAML instead of JSON?
Content-Type: application/x-yaml and format the body in YAML. The API accepts both JSON and YAML for both the schema definition and the full request body.What are the schema limits?
type and description.How does extract/file work?
file_id instead of a video_url. The file must have been uploaded and transcribed first.