Skip to main content
Visual Intelligence is a collection of stateless REST APIs you call directly on your own video, image, and audio data. No persistent library, no indexing — pick the endpoint that matches your task, send a request, get a result. If you want a persistent, indexed video + image library you can query by natural language or image, you want the Visual Search product instead.

​
Core Capabilities

Video Scraping

Metadata, transcripts, captions, and comments from TikTok, YouTube, Instagram, and X/Twitter.

Video Processing

Upload, signed-URL upload, frame extraction, download, delete — manage assets used by other VI APIs.

Video & Image Understanding

Call Gemini / Nova / Qwen / GPT VLMs and ILMs directly. Pre-packaged tasks for frame description and summary.

Audio Transcription

Whisper, ElevenLabs, AssemblyAI providers + speaker diarization / recognition for files and live streams.

Live Stream Processing

Real-time content moderation and prompt-driven understanding on RTMP / RTSP streams.

Embeddings

Image / video / text embeddings for semantic search and retrieval pipelines.

​
Hosts

Most endpoints live on https://mavi-backend.memories.ai/serve/api/v2. Two specialty hosts are used for specific groups: Each endpoint page includes a banner with its exact host and the authentication header to use.

​
Next Steps

API Overview

Browse every endpoint group with one-line descriptions — pick the right one for your use case.

Create Your API Key

Generate a key in under 2 minutes and start calling endpoints.