How Clean Works Without Quality Loss
Most metadata removers open your photo on a canvas and save it as a new file, which recompresses the pixels and hurts image quality. Clean works directly on file bytes instead.
1. The Problem With Standard Image Tools
When you upload a photo to an online converter or edit it in standard software, the application decodes the compressed image data into raw RGB pixels, strips the headers, and runs an encoder to save a new file.
Because JPEG and WebP use lossy compression algorithms, re-encoding causes generational loss: edges blur, subtle color gradients develop banding, and file sizes often increase unnecessarily.
2. Byte-Level Container Parsing
Clean never decodes your image into a pixel canvas. Instead, it reads the binary structure of the container format and separates the metadata blocks from the actual compressed image data:
JPEG (JFIF)
Metadata lives in distinct application marker segments before the start of the image scan (SOS, marker 0xFF 0xDA). EXIF is stored in APP1 (0xFF 0xE1), ICC color profiles in APP2 (0xFF 0xE2), IPTC records in APP13 (0xFF 0xED), and C2PA Content Credentials in APP11 (0xFF 0xEB).
Clean parses each segment sequentially, drops the selected metadata segments, and leaves the compressed entropy scan data (SOS through EOI 0xFF 0xD9) untouched down to the exact bit.
PNG
PNG files consist of an 8-byte signature followed by structured chunks. Image pixels are stored in IDAT chunks, color profiles in iCCP, EXIF in eXIf, and text parameters in tEXt, zTXt, and iTXt.
Clean preserves critical chunks (IHDR, IDAT, PLTE, IEND) and strips ancillary metadata chunks. The decompressed pixels and checksums match the original file bit-for-bit.
WebP
WebP uses the RIFF container format. Pixel data is stored in a VP8 (lossy) or VP8L (lossless) chunk, while metadata is stored in separate EXIF, XMP, or ICCP chunks.
Clean drops the selected metadata chunks, updates the flag bits in the VP8X extended header, and recalculates the 4-byte RIFF container length without touching the VP8 stream.
3. Preserving Ultra HDR and Gain Maps
Modern smartphone photos (such as Ultra HDR JPEG on Android and iPhone HDR photos) contain two separate images inside a single file: a primary standard dynamic range (SDR) image, and a secondary gain map image stored after the primary image end marker.
These images are linked by Multi-Picture Format (MPF) metadata in an APP2 segment and XMP container declarations. Careless metadata strippers delete this XMP, breaking HDR rendering and leaving an orphaned secondary image.
Clean preserves the Ultra HDR gain map structure by default, cleans both primary and secondary images, and recomputes the MPF byte offsets so your photos render with full HDR highlights on compatible displays.
4. Bit-Identical Proof
You can verify that Clean preserves 100% of your image data:
- Take an original camera photo and inspect it with our tool.
- Clean the photo using the "Remove AI markers only" or "Remove private data" preset.
- Extract the entropy data (from the Start of Scan marker to End of Image) from both files using a hex editor or cryptographic hash (SHA-256).
The hash of the image data in the cleaned file matches the original photo exactly. Not a single pixel is touched, recompressed, or altered.
5. Why We Do Not Touch SynthID
C2PA, EXIF, and IPTC are metadata: text, timestamps, coordinates, and cryptographic signatures stored in file headers outside of the pixels.
SynthID (developed by Google DeepMind) is not metadata. It is a statistical watermark embedded directly into the pixel values in the frequency domain. Removing SynthID requires altering the image pixels.
Because Clean is strictly a lossless metadata tool running 100% in your browser, it never alters pixels and does not remove SynthID. We clearly state this so you always know what to expect.
Try It on Your Photos
Your photos stay on your device. Run them through the cleaner and inspect what gets removed.