Image Hash Changer
Online
Change the technical fingerprint of an image by adding invisible pixel-level noise. The image looks identical to your eye; its perceptual hash shifts enough to be treated as a new file by duplicate-detection systems. Runs entirely in your browser.
What this actually does
We load your image into a browser canvas, get the raw RGB pixel values, and add a small amount of Gaussian noise to every channel. The standard deviation is configurable (slider above) but defaults to ±1.5 brightness units out of 255— far below the threshold of human visibility at normal viewing distance.
Because the noise is per-pixel and randomly distributed, average perceptual hashes (pHash, dHash, aHash) that platforms use for duplicate detection and reverse-image search will compute a different value for the output than for the input.
Bonus: metadata also gets stripped
As a side effect of running through the canvas, every metadata block in the original (EXIF, GPS, C2PA, AI tags, XMP) is left behind. So this is effectively perceptual hash change + metadata cleaner in one pass. Output is always PNG (lossless), so the noise survives intact.
Be specific about what this can and can’t do
- Can: shift perceptual hashes (pHash, dHash, aHash) by enough to defeat naïve duplicate detection.
- Can: strip embedded metadata in the same pass.
- Cannot: defeat content-based AI detectors that look at pixel statistics or learned features — those don’t rely on hashes.
- Cannot: remove pixel-level watermarks like Google’s SynthID — those are designed to survive low-amplitude noise.
- Cannot: prevent identical-pixel reverse search if the original is already public on the web — Google can match by exact bytes too.
Tuning the noise
σ = 0.5–1.5 is invisible. Below ~1.0 some perceptual hashes may not shift far enough — try higher if your tests still match. σ = 2.0–2.5 is “very subtle” — only visible on flat areas at high zoom. σ = 3.0 reads as faint film grain; avoid unless you need maximum hash shift.
Frequently asked.
Is this legal?+
Adding noise to your own image is just editing it — completely fine. Using it to defeat copyright protection on someone else's image, evade a court-ordered hash block, or circumvent CSAM detection is a different question and generally not legal anywhere. We built this for the legitimate cases: re-uploading your own work that a platform incorrectly flagged as a duplicate, sharing variants of marketing images, defeating overly aggressive deduplication.
Why does the output have to be PNG?+
JPEG compression quantizes pixel values in 8×8 blocks. The Gaussian noise we add is mostly removed by that quantization, defeating the purpose. PNG is lossless — every pixel value survives exactly as we wrote it.
Will the result work for evading [specific platform's] detector?+
We can't promise that. Modern platforms use multiple detection layers: perceptual hashes, content fingerprints, learned image embeddings, OCR, face recognition. This tool shifts perceptual hashes — one input among many. Test on a low-stakes account first.
Does the file get uploaded?+
No. The whole pipeline runs in your browser via the Canvas API. Open dev tools, watch the Network tab while you drop a file — zero requests fire. The page works fully offline once loaded.