PixelSeal
Experiment8 min · February 28, 2026

Torture Test v3: PixelSeal Survival Under 53 Real-World Transforms

We subjected PixelSeal-sealed images to 53 destructive transforms — JPEG recompression, aggressive resizing, cropping, rotation, and platform-simulated processing. 26 verified, 27 detected, 0 timeouts.

watermarkingrobustnessjpegcompressionresize

Experiment Setup

We generated a test image sealed with PixelSeal V2 (RS(63,28) with 9-tile DCT embedding) and ran it through our Torture Test v3 suite — 53 automated transforms covering JPEG quality sweeps (Q20–Q95), resize chains (1080p, 720p, 75%, 50%), crop factors (5%, 10%, 20%, 30%), rotations, and combined multi-step degradations.

🔬Method
This experiment uses PixelSeal's standard embed + verify pipeline. You can reproduce results using the developer tools.

Methodology

Each transform is applied independently to the sealed source image using Sharp. The transformed image is then verified using PixelSeal's two-stage budgeted scale search (Phase 1–5) with a 60-second timeout per candidate. Results are classified as VERIFIED (valid HMAC + CRC), DETECTED (sync found but signature invalid), or TIMEOUT.

Results Summary

Out of 53 transforms: 26 produced VERIFIED results (full cryptographic validation), 27 were DETECTED (watermark structure found, but too much corruption for HMAC validation), and 0 timed out. This represents a major improvement from v2 testing where 14 cases timed out. The budgeted search algorithm reduced worst-case verification time from >180s to <60s.

Key Findings

JPEG recompression down to Q60 is fully recovered. Resize to 75% of original dimensions is recovered with RS error correction (4 symbols corrected). Combined transforms (resize + JPEG) show graceful degradation. The sync marker detection rate was 100% across all transforms, meaning the watermark structure is always found — only the payload integrity varies.

💡Key Finding
These results demonstrate that PixelSeal watermarks are production-ready for workflows where images undergo typical social media processing (JPEG recompression + resize). The two-stage budgeted search...

Implications for Production

These results demonstrate that PixelSeal watermarks are production-ready for workflows where images undergo typical social media processing (JPEG recompression + resize). The two-stage budgeted search ensures verification completes within practical time bounds even for heavily degraded images.

Reproduce this experiment

Seal your own image and verify it after applying transforms. The same tools we used are available to you right now.

Frequently Asked Questions

What does 'DETECTED but not VERIFIED' mean?

Why did some resize transforms fail verification?

How does the budgeted search prevent timeouts?