ingest-content

Use when building a media ingestion pipeline — processing raw video/audio/image files, extracting technical metadata, AI-driven scene classification, deduplication by perceptual hash, or organizing footage into an indexed library. Also use when setting up batch processing for large media collections or building a RAG-ready asset database.

Model	Source
sonnet	pack: video

Full Reference

┏━ 🔧 ingest-content ━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ ┃ Use when building a media ingestion pipeline —… ┃ ┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛

Content Ingestion Pipeline

Systematic pipeline for transforming raw media files into a structured, searchable, AI-enriched library. Five stages: intake → metadata extraction → AI classification → deduplication → indexing. Each stage is independently composable.

Quick Reference

Item	Value
Supported video formats	.mp4, .mov, .avi, .mkv, .webm, .mxf
Supported audio formats	.mp3, .wav, .aac, .flac, .m4a
Supported image formats	.jpg, .jpeg, .png, .webp, .tiff, .raw
Metadata tool	fluent-ffmpeg + @ffprobe-installer/ffprobe
Image processing	sharp
AI classification model	gemini-2.5-flash via @google/genai
Dedup algorithm	dHash (difference hash) + Hamming distance
Near-duplicate threshold	Hamming distance < 5
Batch concurrency	p-queue, default 4
Database	Postgres + pgvector (HNSW index for embeddings)

Reference Index

I want to…	File
Scan directories, validate file types, or extract video/image metadata with ffprobe/sharp	intake-and-metadata.md
Classify scene type, mood, and content tags from a video keyframe using Gemini Vision	ai-classification.md
Compute perceptual hashes, detect near-duplicates, or compare with Hamming distance	deduplication.md
Run the full batch pipeline, set up the Postgres schema, or write the library index	pipeline-and-indexing.md

Usage: Read the reference file matching your current task from the index above. Each file is self-contained with code examples and inline gotchas.