We launched the 100% in-browser 5-Tier ISOBMFF (MP4, MOV, M4V) and EBML (WebM, MKV) Video Sanitization Engine. Creators can now strip C2PA JUMBF UUID manifests, recalculate 32-bit (stco) and 64-bit (co64) chunk offset tables with zero video re-encoding, scrub H.264/HEVC SEI NAL unit generator strings (OpenAI Sora, Runway Gen-3 Alpha, Kling AI, Luma Dream Machine, Pika Labs, Hailuo MiniMax, Google Veo, CapCut), and synthesize authentic cinema camera hardware atoms directly in browser RAM.
Recursive ISOBMFF atom tree parser excising C2PA JUMBF UUID boxes (d8fec3d6-1b0e-483c-9297-5828877ec481) and Adobe XMP UUID boxes.
Exact 32-bit (stco) and 64-bit (co64) chunk offset table recalculation so fast-start MP4 and MOV videos remain 100% playable and frame-synced with zero re-encoding.
Length-preserving in-place H.264/HEVC SEI NAL unit and EBML tag scrubbing for Sora, Runway, Kling, Luma, Pika, Hailuo, Veo, SynthID, and CapCut.
Authentic QuickTime udta/meta/ilst hardware synthesis for Apple iPhone 16 Pro Max, Sony ILME-FX3, Canon EOS R5 C, ARRI ALEXA Mini LF, RED V-RAPTOR 8K, DJI Mavic 3 Pro Cine, and DJI Osmo Pocket 3.
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 normalize authentic camera hardware EXIF data—specifically matching standard device profiles. 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.
Normalized standard optical parameters (Focal length, F-stop, ISO, Exposure time, Color space).
Added secondary profile for mobile main camera.
100% binary-level buffer rewrite with zero server storage.
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.