We updated the UnlabelAI core engine with edge-preserving bilateral micro-smoothing and sub-pixel coordinate desynchronization to counter Google DeepMind SynthID and OpenAI ChatGPT multi-layer provenance measures. The engine now completely neutralizes both signed C2PA container manifests and pixel-embedded SynthID watermarks verifiable by OpenAI Verify and Google SynthID Detector.
We upgraded our in-browser metadata engine to inject authentic camera hardware EXIF data—specifically simulating a Sony A7 IV with a 35mm F1.4 GM lens and an Apple iPhone 15 Pro. When social platforms scan media, images with blank headers often trigger spam evasion heuristics; authentic optical metadata signals genuine physical capture to platform ingest algorithms.
Injected Sony A7 IV optical parameters (Focal length, F-stop, ISO, Exposure time, Color space).
Added secondary profile for Apple iPhone 15 Pro main camera (24mm f/1.78).
100% binary-level buffer rewrite in browser RAM using piexifjs.
Standard metadata strippers only delete header tags while leaving invisible AI diffusion pixel patterns intact. We deployed a sub-perceptual phase jitter algorithm that disrupts Google DeepMind SynthID frequency harmonics without altering visual clarity, preventing automated visual classifiers on Meta and TikTok from applying retroactive AI Info badges.
Sub-perceptual ±1 RGB perturbation applied across pseudo-random pixel coordinate grids.
High-volume creators and social media agencies can now drag-and-drop up to 50 images in a single dropzone session. All files are scrubbed in parallel using client-side Web Workers and packaged into an uncompressed ZIP archive for instant 1-click download with zero network latency.
Multi-threaded in-browser sanitization pipeline.
Client-side ZIP compression using JSZip.
Individual and bulk download controls with progress indicators.
Our initial release established 100% client-side sanitization using the HTML5 Canvas API. By drawing raw RGB arrays onto an isolated in-memory canvas context, all provenance manifests, Adobe Content Credentials, Midjourney generator tags, and IPTC edit histories are eliminated before your image touches a social network.
Zero server upload architecture ensures complete creator privacy.
Full support for JPEG, PNG (lossless), and WebP formats.
Instant drag-and-drop workflow with live before/after inspector.