AI Metadata Inspector & Cleaner
See what your image is telling machines.
Inspect EXIF, XMP, IPTC, C2PA, JUMBF and AI-generation parameters directly in your browser. Lossless binary inspection and surgical removal without re-encoding.
or browse local filesystem
Running client-side binary parsing inside isolated Web Worker sandbox.
The file format could not be safely parsed or contains unsupported binary structures.
* Note: Pixel-level watermarking and visual pattern recognition operate in the spatial or frequency domain of the image pixels, not in container metadata. This tool operates exclusively on binary container segments without altering pixel data.
▶RAW METADATA INSPECTION PAYLOAD (JSON)[ Click to expand ]
Choose container mutation strategy
All operations perform surgical binary stream editing on the original container without pixel decompression or re-encoding.
Local Verification Matrix
Image data preserved byte-for-byte
Verified exact CRC32 across image-bearing frame payload without decompression or re-encoding.
| Metadata Component | Original File | Cleaned Output | Status |
|---|
How the Local Binary Engine Works
Scans APP markers, PNG chunks, RIFF boxes.
Distinguishes C2PA, JUMBF, EXIF, AI tags.
User selects targeted mutation mode.
Removes segments without canvas re-encode.
Validates bitstream CRC32 invariance.
Signal Classification & Removal Map
Standardized overview of supported metadata structures versus pixel-level watermarks.
| Signal Type | Container Mechanism | Removal Capability |
|---|---|---|
| EXIF | TIFF IFD headers in APP1 / eXIf | Supported (in full strip mode) |
| XMP | RDF XML payload in APP1 / iTXt | Supported (AI schemas in AI-only mode) |
| IPTC | 8BIM Photoshop records in APP13 | Supported (in full strip mode) |
| C2PA | Cryptographic manifest store | Supported where a valid C2PA manifest is recognized |
| JUMBF | ISO/IEC 19566-5 box container | Supported where safely identified as removable C2PA/provenance data |
| AI Prompts | A1111 / ComfyUI tEXt/iTXt strings | Supported where recognized |
| SynthID | Frequency-domain pixel watermark | Outside Scope (Pixel-level signal) |
| Visual AI Detection | Neural visual classifier inference | Outside Scope (Model inference) |
Under the Hood
Why No Canvas or HTML ImageBitmap?
Standard web utilities re-encode images by drawing them to an HTML <canvas> and exporting via toBlob('image/jpeg'). This re-compresses image pixels, changes chroma subsampling, and introduces generational loss. This tool never decodes pixels — it parses and filters the container bitstream directly.
Fail-Closed Safety Guardrail
If an image bitstream contains corrupted framing, unknown critical chunks, or unrecognized JUMBF superbox layouts that cannot be safely parsed, the engine fails closed with an explicit descriptive error instead of guessing or risking binary corruption.
JUMBF vs. C2PA Distinction
JUMBF (ISO/IEC 19566-5) is a general box-structured binary container, whereas C2PA (Content Credentials) is an application standard built on top of JUMBF. The engine inspects the JUMBF description box (jumd) to specifically target provenance payloads.
Web Worker Sandboxing
All parsing and bitstream filtering runs within a sandboxed Web Worker thread. Large images (up to 50MB) are processed without blocking the main UI thread. Memory is released immediately via ArrayBuffer deallocation.