How It Works
1
Upload
Send videos or images to the API. Each is indexed automatically and stored in the appropriate library.
2
Auto-Index
Every upload is processed through two pipelines in parallel:
- Embedding — visual frames and audio are encoded into semantic vectors, enabling natural-language and image search
- Caption & transcription — the platform generates descriptions and transcribes spoken audio, enabling text-level search
UNPARSE (processing) to PARSE (searchable).3
Search
Query your indexed content by natural language, image, or transcript keyword. Results are ranked by relevance.
Three Libraries
Visual Search has three distinct libraries. Understanding which library holds what content is essential — uploads go to different libraries, and each library has its own search endpoints.
Private Video Library and Private Image Library are separate. Uploading a video does not add it to your image library, and uploading an image does not add it to your video library. They have different upload endpoints, different search endpoints, and return different result formats.
Search Modalities
Once indexed, content is searchable in two fundamental ways:Next Steps
Get Your API Key
Generate a key to authenticate all API requests.
Upload Content
Upload videos and images — how the indexing pipeline works and which library each upload lands in.
Search
Explore all search endpoints across all three libraries.