Re-upload detection
for social platforms
Banned images and videos reappear minutes after removal — re-encoded, cropped, or AI-upscaled to slip past exact-match filters. At millions of uploads per day, manual review cannot keep up.
← All use casesThe problem
- ✗Millions of uploads per day make manual re-review impossible — enforcement must be automated.
- ✗Bad actors re-encode video and re-save images specifically to defeat SHA-256 hash blocklists.
- ✗CSAM-adjacent and harassment content keeps circulating because pixel-level edits invalidate existing fingerprints.
- ✗Every hour of undetected re-distribution increases legal exposure and erodes user trust.
73%
of re-uploads happen within 24 hours of original removal
12×
more re-upload attempts per removed piece of content vs 3 years ago
< 1%
of re-uploads are byte-identical — the rest evade SHA-256
How Rechase helps
Sub-50ms at any scale
Three detection layers run in parallel — SHA-256, perceptual hashing, and CLIP semantic embeddings — returning a verdict before the upload reaches storage. No throughput ceiling.
Survives every re-encode
CLIP embeddings capture semantic content, not pixels. A video re-encoded to H.265, colour-graded, and letterboxed still matches the original with 90%+ confidence.
Audit trail for legal teams
Every block is logged with fingerprint scores, detection method, and a link to the original removal order. Exportable for CSAM reporting and regulatory compliance.
Detection methods
What each layer catches specifically on social platforms.
| Method | What it catches on social platforms | Latency |
|---|---|---|
| SHA-256 | Identical re-posts — users who copy-paste files directly without editing | < 1ms |
| pHash | Cropped screenshots, watermark removals, aspect-ratio changes, colour filter additions | < 5ms |
| CLIP | Re-encoded videos, AI upscales, deepfake variants, style-transferred images with same harmful content | 30–50ms |
Ready to protect your
social platform?
30-minute call. We walk through your upload volume, existing hash blocklist, and estimate time-to-integration.
← All use cases