Image Metadata & AI Watermark Cleaner
Strip EXIF, GPS, XMP, C2PA Content Credentials, and embedded AI workflows โ plus disrupt in-pixel steganography (SynthID) with sub-pixel micro-warping and DCT re-quantization.
Image Metadata, C2PA & AI Provenance Cleaner (Single & Batch)
Drop one or multiple images to inspect embedded metadata first. Strips 100% of OpenAI / ChatGPT C2PA Content Credentials (LinkedIn CR badge), EXIF, GPS, XMP, and ComfyUI workflows with zero visual quality loss.
Who Uses This Tool
Listing photos taken at home can carry your exact GPS coordinates and phone model. Clean the whole batch and upload the ZIP.
Remove location, device serials and author fields before sharing images from a source or a sensitive location.
Forums, classifieds and some websites keep EXIF intact. Check what a photo reveals, then strip it.
Prompts, seeds and ComfyUI workflows are often embedded in PNG files. Remove them before you share an image.
Use it responsibly. Removing location and device data protects your privacy. Some platforms and regulations require AI-generated content to stay labeled, so do not use this tool to pass off AI images as real photographs where disclosure is required.
Last tested: in Chrome, Firefox, Safari and Edge (current versions).
More Free Browser Diagnostic & Utility Tools
Full 104-key test with live press duration & latency measurement.
Check polling rate up to 8000Hz, jitter, click latency & heatmap.
Real Hz, frame consistency grade, motion clarity & VRR simulator.
Check for dead pixels, backlight bleed, color banding & uniformity.
Live camera FPS, mic waveform with clipping check & stereo test.
How It Works โ And What Browser Scrubbing Can & Cannot Do
Most "watermark removers" claim to erase every trace of AI generation without explaining the difference between file headers and pixel steganography. Here is exactly what happens to your file in the browser:
| Data / Watermark Type | Examples | What This Tool Does | Result |
|---|---|---|---|
| Container Metadata | EXIF, GPS coordinates, camera serials, XMP, IPTC | Lossless mode deletes the metadata segments directly. Re-encode mode decodes the pixel bitmap and writes a brand-new file container. | 100% Removed |
| C2PA / Content Credentials | JUMBF manifests (c2pa, caBX) from ChatGPT, DALL-E, Firefly | Drops all APP11/JUMBF, caBX and WebP C2PA boxes, then re-scans the output to confirm | 100% Removed |
| AI Workflow Chunks | ComfyUI JSON graphs, Automatic1111 prompts, PNG tEXt/iTXt | Strips all ancillary PNG/WebP text chunks and verifies binary output | 100% Removed |
| Frequency Watermarks | Google SynthID, Stable Diffusion invisible-watermark | Deep Scrub applies 0.25% sub-pixel scale, sinusoidal wave warp, edge-adaptive dither & 8x8 DCT pass. Lossless mode does not touch pixels. | Degraded / Disrupted (best effort, not verifiable) |
| Latent / Structural Watermarks | Tree-Ring watermarks, visual AI classifier artifacts | Cannot be removed by pixel math alone (requires AI VAE regeneration) | Not Removed |
Frequently Asked Questions
Does this tool upload my images to a server?
No. All decoding, binary inspection, canvas redrawing, ZIP building and frequency dithering happen 100% locally inside your browser using the HTML5 Canvas and Web Crypto/TypedArray APIs. Your images never leave your device.
What is the difference between C2PA metadata and SynthID?
C2PA Content Credentials and EXIF tags are stored in the file container headers and are removed when the file is cleaned. SynthID is a steganographic watermark embedded into pixel frequency values, requiring sub-pixel warping, edge-adaptive dithering and DCT quantization to disrupt.
Will Deep Scrub mode ruin my image quality?
No. Smooth areas (sky, skin, gradients) are altered by only ยฑ1 RGB value out of 255, while stronger frequency disruption is masked inside complex textures like hair and foliage. If you need 100% bit-exact pixels, switch to Metadata Only mode or use the Lossless strip mode in the batch cleaner.
What is Lossless strip mode and when should I use it?
Lossless strip removes EXIF, GPS, XMP, C2PA, IPTC, comments and PNG/WebP text chunks directly from the file without decoding the picture. Pixels, JPEG compression and the ICC color profile stay exactly as they were, so there is no quality loss and no color shift. Use it for privacy cleaning. It does not disrupt pixel-level watermarks such as SynthID.
Which formats are supported, and what happens to HEIC and AVIF?
JPEG, PNG and WebP can be cleaned losslessly or re-encoded. HEIC/HEIF and AVIF files are scanned for metadata and then converted to JPEG or PNG, because browsers cannot save images in those two formats. The HEIC/AVIF metadata report is a best-effort scan.
Why isn't a simple PNG redraw enough to remove SynthID?
Redrawing an image onto an HTML5 canvas and saving as a lossless PNG keeps the exact RGB pixel values intact. Because SynthID embeds its signal directly into pixel frequency coefficients rather than file headers, a 1:1 lossless redraw preserves the watermark. Switching to Deep Scrub shifts the spatial grid by 0.25%, applies non-linear wave warping, injects edge-masked noise, and forces an 8x8 DCT quantization pass.
Does Deep Scrub flatten transparent PNG backgrounds?
No. Before running the RGB channels through the in-memory JPEG DCT quantization pass, the tool copies your original Alpha channel into a separate buffer and restores it pixel-for-pixel before generating your final PNG or WebP file.
What is the 20x Pixel Diff Map toggle?
Because Deep Scrub changes pixel values by only ยฑ1 to ยฑ3.4 out of 255, the changes are invisible to the naked eye. Checking "Show 20x Pixel Diff Map" multiplies the mathematical difference between the original and cleaned pixels by 20 so you can verify that edge-adaptive frequency scrubbing took place.