Larc adds an imperceptible perturbation to your photos that disrupts perceptual hashes, CLIP embeddings, and reverse-image indexes — while keeping the image visually near-identical.
Free. No sign-up needed to try. PNG & JPG up to 20 MB.
How it works
Larc doesn't watermark, doesn't blur, doesn't distort. It shifts the fingerprints that search systems use to match images.
Drop a PNG or JPG. Everything runs locally in your browser — the file never touches a server.
Larc applies a sub-pixel warp, adaptive high-frequency noise, and a chroma micro-shift keyed to texture in the image.
Get a near-identical file with a fresh hash, disrupted embeddings, and stripped metadata.
What Larc disrupts
Adversarial image cloaking is real, but no tool can promise zero results on every search engine. Here's what Larc consistently disrupts — and what it doesn't.
| Signal | Effect | Notes |
|---|---|---|
| Perceptual hashes (pHash, dHash, aHash) | Strong | Reliably shifts hash by many bits |
| CLIP / OpenCLIP embedding similarity | Strong | Meaningful cosine-distance drop |
| SIFT / ORB keypoint matching | Strong | Sub-pixel warp shifts keypoints |
| EXIF / XMP / ICC metadata fingerprints | Strong | Stripped entirely on export |
| File hash (MD5/SHA) | Strong | Fresh bytes every export |
| Google Lens end-to-end | Variable | Variable — Google uses updating ensembles |
| Human recognition | Variable | Image still looks the same to you |
Try it now
Everything happens in your browser. Nothing uploads. See the before/after and the SSIM score in real time.
FAQ
No — and anyone who promises that is lying. Google Lens uses a moving ensemble of models. Larc reliably disrupts perceptual hashes, embedding similarity, and keypoint matching, which are the signals most reverse-image systems (including many of Google's) rely on. Effectiveness against Google specifically varies per image.
Almost never. Larc concentrates changes in textured regions where the human eye is least sensitive, and reports a numeric SSIM score against the original so you can pick a strength that keeps the image visually intact.
No. The entire pipeline runs in your browser using Canvas2D. Your original never leaves your device unless you explicitly save the cloaked output to history.
PNG and JPG, up to 20 MB. Output preserves the input format by default, or you can force JPG or PNG.
Yes. Larc doesn't hide the image from anyone looking at it — it makes automated fingerprinting less reliable, similar to Glaze or PhotoGuard. Don't use it to evade legitimate content moderation.