Part of the Abstract Media Intelligence Platform
This module handles video ingestion and multimodal extraction within a unified media pipeline.
abstract_videos processes:
- video download + metadata registry
- transcription (Whisper) + frame OCR
- NLP enrichment and structured storage
Full system: https://github.com/AbstractEndeavors/abstract-media-intelligence
abstract_videos — Video Processing & Media Intelligence Pipeline
A structured pipeline for transforming video content into searchable, metadata-rich, and SEO-optimized assets, combining ingestion, transcription, OCR, NLP enrichment, and persistent storage.
Designed for:
- large-scale video ingestion
- transcription and content extraction
- media indexing and search
- automated metadata generation and SEO
🔹 What This System Is
abstract_videos is not a downloader or transcription tool — it is a multi-stage media processing system:
- ingests video from URLs or local sources
- extracts audio, frames, and text
- performs transcription (Whisper)
- applies OCR to extracted frames
- enriches content via NLP (keywords, summaries, titles)
- persists structured results to database or filesystem
The system produces fully structured video representations usable for:
- search
- indexing
- content generation
- analytics
🔹 Pipeline Overview
Video Input (URL / File)
↓
Download + Registry (yt-dlp + metadata)
↓
Video Processing
├─ Conversion / normalization
├─ Audio extraction
├─ Frame extraction
↓
Content Extraction
├─ Transcription (Whisper)
├─ OCR on frames
↓
NLP Enrichment
├─ Summarization
├─ Keyword extraction
├─ Title generation
↓
Metadata Assembly
↓
Persistence Layer
├─ Database (JSONB structured storage)
└─ Filesystem (artifacts + media)
Pipeline
flowchart TD
A[Video URL / Local Video]
B[VideoDownloader + Registry]
C[Normalization / Conversion]
D[Audio Extraction]
E[Frame Extraction]
F[Whisper Transcription]
G[Frame OCR]
H[NLP Enrichment\nSummary + Keywords + Title]
I[Metadata Assembly\nCategory + Thumbnail + SEO]
J[Persistence Layer\nFilesystem + DB / JSONB]
K[Searchable / Structured Video Record]
A --> B --> C
C --> D --> F
C --> E --> G
F --> H
G --> H
H --> I --> J --> K
🔹 Core Capabilities
Video Ingestion & Registry
- URL normalization and ID generation
- Metadata extraction via yt-dlp
- Persistent registry with atomic updates and locking
Processing Pipeline
- Video normalization and format handling
- Audio extraction for transcription
- Frame extraction for visual analysis
Transcription & OCR
- Whisper-based transcription pipeline
- Frame-level OCR for embedded text
- Combined multimodal text extraction
NLP & Metadata Enrichment
- Keyword extraction and refinement
- Title generation from summaries
- Category inference based on content
- Thumbnail selection via frame sharpness analysis
Structured Persistence
-
PostgreSQL storage with JSONB fields for:
- raw info
- metadata
- transcripts
- captions
- thumbnails
- aggregated outputs
-
Upsert-based lifecycle management for idempotent processing
🔹 Dual Pipeline Model (Key Concept)
The system supports two execution modes:
1. Local / Read-Write Pipeline
- full processing on local machine
- filesystem-based outputs
- direct artifact generation
2. Database-Centric Pipeline
- persistent storage as primary interface
- JSONB-backed structured data
- incremental updates and enrichment
🔹 Design Intent
Database + local modules as primary External ML (HuggingFace / APIs) as secondary
This enables:
- offline-first operation
- reproducibility
- plug-and-play ML upgrades
🔹 Architecture
Core Components
-
VideoDownloader
- ingestion + metadata acquisition
-
infoRegistry
- centralized state + persistence
-
VideoTextPipeline
- orchestrates processing stages
-
Metadata Console
- post-processing and optimization (summaries, SEO)
-
Database Layer
- structured storage with upsert semantics
🔹 Key Design Decisions
Idempotent Processing
- all steps tracked via
processed_steps - pipeline resumes without duplication
- safe reprocessing of partial runs
Structured Over Raw
Everything is stored as structured JSON:
- transcripts
- keywords
- metadata
- derived content
Multimodal Extraction
Combines:
- audio → text (transcription)
- image → text (OCR)
- text → meaning (NLP)
Registry as Source of Truth
- central video registry
- thread-safe and process-safe updates
- ensures consistency across runs
🔹 Why This Exists
Most video pipelines:
- stop at transcription
- lack structure
- are not searchable
- are not reusable
abstract_videos transforms video into:
- structured data
- searchable content
- SEO-ready metadata
- indexable media assets
🔹 Example Use Cases
- video → searchable content pipelines
- media indexing platforms
- transcription + analytics systems
- SEO content generation
- LLM-ready dataset creation
🔹 Integration Context
This system integrates directly with:
abstract_hugpy→ NLP / summarization / keyword extractionabstract_ocr→ image/frame OCRabstract_pdfs→ document pipeline
🔹 Design Philosophy
- Media is data, not just content
- Structure enables reuse
- Pipelines should be resumable and deterministic
- Local-first, cloud-optional
Metadata
Release files for abstract-videos 0.0.0.285
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| abstract_videos-0.0.0.285.tar.gz | 21.5 kB | Without provenance | Details for abstract_videos-0.0.0.285.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| abstract_videos-0.0.0.285-py3-none-any.whl | Python 3 | none | any | Details for abstract_videos-0.0.0.285-py3-none-any.whl |
Total release size: 43.8 kB
Release files / abstract_videos-0.0.0.285.tar.gz
| Download URL | abstract_videos-0.0.0.285.tar.gz |
|---|---|
| Size | 21.5 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
8ace8ac7e8e31e17f2726f67e20522b028d57eb626ba1c4533f3d1b67e76107a
|
|
BLAKE2b-256 checksum How to use checksums |
cee365280c36b37d74d5f6f51b79f1a5b3e4d22a01cc897a280e2a9dbab1e278
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.12.3
|
Release files / abstract_videos-0.0.0.285-py3-none-any.whl
| Download URL | abstract_videos-0.0.0.285-py3-none-any.whl |
|---|---|
| Size | 22.3 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
937babedb6f03c341de2073257bd9e8a66c42d0929bba5ecd1549b8ca7755a49
|
|
BLAKE2b-256 checksum How to use checksums |
77fe869cc50a40139b70fd134e8cd199c824ca38bb37b7cd55693a448f3bedba
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.12.3
|
Security for abstract-videos 0.0.0.285
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.