MetaGhost Team
Content fingerprinting researchers. We reverse-engineer platform detection so creators don't have to.
Who we are
MetaGhost is a small, independent team of engineers and ML researchers obsessed with one question: how do platforms actually detect that a video is a repost?
The answer is more interesting than the marketing-speak you'll find elsewhere. It's not just "watermarking". It's an ensemble of perceptual hashes, neural embedding models (Meta's SSCD, ByteDance's 灵识系统), video fingerprinting pipelines (TMK+PDQF), and metadata heuristics. Each platform runs a different stack, with different thresholds, at a billion-query scale.
We read the papers, benchmark the systems, and ship the tools. Everything we build runs 100% on the user's machine: no cloud, no uploads, no data harvesting. The software we ship is the same software we use ourselves to understand how these systems behave.
What we focus on
Research
White-box benchmarks of content fingerprinting models. PyTorch + TF.js validation. Cross-platform threshold mapping from published papers and empirical testing.
Engineering
Native Electron app with embedded TensorFlow.js, FFmpeg, yt-dlp. Adversarial pipeline running PGD attacks against an SSCD+EfficientNet+nsfwjs ensemble at full resolution.
Education
Technical articles explaining how detection actually works, with sources, real thresholds, and reproducible examples. No fluff, no clickbait.
Product
MetaGhost desktop for Windows & macOS. Batch processing, metadata injection, adversarial perturbations, similarity detection, short-maker templates. Ships monthly.
Areas of expertise
Content Fingerprinting
Meta SSCD, YouTube Content ID, TMK+PDQF, ByteDance 灵识
Perceptual Hashing
pHash, dHash, aHash: thresholds and bypass vectors
Adversarial ML
PGD, TI/DI-FGSM, ensemble attacks on embedding models
Metadata Engineering
EXIF, IPTC, XMP injection with realistic device signatures
Platform Detection
TikTok, Instagram, YouTube, Snapchat shadowban mechanics
Video Pipeline
Per-frame PGD, FFmpeg filter graphs, CRF-robust perturbations
Published research
Our technical writing on detection systems. Grounded in published papers, not marketing.
Get in touch
Questions about our research, the product, partnership inquiries, or press? We read every message.
Message us on Telegram