How to Strip EXIF, XMP, and Hidden Thumbnail Metadata from Images in Python (and Browser)

When users upload profile avatars, marketplace photos, or bug-report screenshots to a web application, or when you publish images to a documentation repo or CMS, those files carry up to 150KB of hidden binary metadata inside their container headers.
Most developers know about GPSLatitude and GPSLongitude inside the standard EXIF APP1 segment. Fewer realize that modern JPEG, PNG, TIFF, and WebP files also carry:

1. Hardware Serial Numbers (BodySerialNumber, InternalSerialNumber): Globally unique identifiers that link anonymous uploads across separate accounts back to a single physical device.
2. XMP Edit Histories & Local OS Paths (XMP:History, DocumentID): XML blocks injected by Photoshop, Lightroom, and OS screenshot tools that can leak local filesystem paths (including macOS/Windows usernames).
3. The "Exifbleed" Thumbnail (IFD1:ThumbnailImage): A secondary 160x120 JPEG preview stored inside the EXIF header. When a user crops out sensitive information (an API key on a monitor, a street sign, or a credit card) using an editor that fails to regenerate the IFD1 preview, the original uncropped photo remains intact inside the metadata header. Here is why tag-blacklisting fails and how to implement a 100% bulletproof Clean-Canvas Pixel Reconstruction pipeline in Python (or use our free browser utility for instant batch scrubbing). --- ## Why Tag Blacklisting Is an Anti-Pattern A common approach to stripping metadata is opening an image library, iterating over known EXIF tags, and deleting keys like 0x8825 (GPSInfo). This fails in production for three reasons:
4. Multi-Container Redundancy: Location and author strings are often duplicated across EXIF (APP1), XMP (XML packet inside APP1), and IPTC (APP13 Photoshop IRB). Deleting the EXIF GPS IFD leaves XMP:GPSLatitude untouched.
5. Proprietary MakerNotes: Apple, Samsung, Canon, and Sony store undocumented binary structs inside MakerNotes (including face-detection coordinates, burst UUIDs, and configured owner names).

* Format Fragmentation: PNG stores metadata in eXIf, tEXt, zTXt, and iTXt chunks, while WebP wraps RIFF EXIF and XMP chunks.

The Solution: Clean-Canvas Pixel Reconstruction in Python (Pillow)

Instead of parsing and mutating untrusted binary header segments, the cleanest security boundary is to extract only the decoded pixel matrix (img.getdata()) and paint those raw RGB/RGBA values onto a brand-new, empty canvas (Image.new(img.mode, img.size)).
Because the output file is constructed strictly from raw pixel values, zero header segments (APP1 through APP15, COM, eXIf, tEXt, or IFD1 thumbnails) are copied over:

import os
from PIL import Image
def strip_all_image_metadata(input_path: str, output_path: str) -> dict:
    """
    Reconstructs an image onto a clean canvas from raw pixel values alone.
    Guarantees 100% removal of EXIF, XMP, IPTC, MakerNotes, and IFD1 thumbnails.
    """
    with Image.open(input_path) as img:
        # Extract raw pixel tuples without any header/container dictionaries
        pixels = list(img.getdata())
        clean_img = Image.new(img.mode, img.size)
        clean_img.putdata(pixels)
        fmt = (img.format or "JPEG").upper()
        if fmt == "JPEG":
            clean_img.save(output_path, format="JPEG", quality=95)
        elif fmt == "WEBP":
            clean_img.save(output_path, format="WEBP", quality=95)
        elif fmt == "PNG":
            clean_img.save(output_path, format="PNG")
        else:
            clean_img.save(output_path, format=fmt)
    return {
        "original_bytes": os.path.getsize(input_path),
        "cleaned_bytes": os.path.getsize(output_path),
    }

Bonus Benefit: Core Web Vitals & LCP Reduction

Camera EXIF blocks, embedded IFD1 preview JPEGs, and bloated Adobe APP13 / ICC color profile headers routinely add 30 KB to 120 KB per image. Stripping them before converting to WebP shaves hundreds of kilobytes off image-heavy pages with zero visual change.

Command-Line Verification with ExifTool

To verify in CI/CD or locally that an image contains zero hidden metadata tags, run Phil Harvey's exiftool:

# Inspect all metadata (including duplicate and unknown tags) grouped by container:
exiftool -a -u -g1 sample_photo.jpg
# Strip all metadata via CLI:
exiftool -all= -o cleaned_photo.jpg sample_photo.jpg

---

Instant Browser Alternative (Batch Process up to 20 Images Free)

If you are on a locked-down machine, mobile device, or just want to drag and drop a batch of photos before pushing a PR or publishing a blog post, we deployed this exact Clean-Canvas pipeline in our free web utility:

• Live Tool: UtilifyAI Free EXIF Metadata Remover (Batch process up to 20 JPEG, PNG, TIFF, or WebP files into a clean ZIP with zero sign-up).
• Companion Image Optimizer: Free PNG to WebP Batch Converter & Free AI Background Remover.
• Full 2026 Deep-Dive & Platform Audit Matrix: How to Remove EXIF Data from Photos: What Is EXIF Metadata & Free Privacy Scrubber Guide (2026).