Watermark images like a pro using Python + Wand + librsvg

This is a Python 2.7 code recipe for applying a semi-transparent vector (SVG) watermark with overlay blending like so:

Source: photo of Fundy Park by Nicolas Raymond.

What’s special about it:

• The watermark is a vector image (SVG), which means it’d look sharp when placed on images of any sizes. The function scales the watermark proportionally to the width of the source image.
• Supports blending modes (what’s that?), including my favorite “overlay”.
• Watermark transparency level is a function argument. SVG transparency is also supported, so different parts of the watermark can have varying opaqueness.

Prerequisites: Wand, imagemagick (for Wand) and librsvg:

sudo apt-get/yum install librsvg imagemagick

If you’re on OS X and use Homebrew: brew install librsvg imagemagick

Install Wand using pip: pip install wand

Make sure rsvg-convert command is available in your shell.

import subprocess

from wand.image import Image as WandImage

def watermark(source, target, watermark, blend_mode='overlay', ratio=1, transparency=0.3):
    source_obj = WandImage(filename=source)
    watermark_obj = WandImage(filename=watermark)
    
    if ratio > 1:
        ratio = 1
    
    watermark_width = int(source_obj.size[0] * ratio)
    watermark_height = int(float(watermark_width) / float(watermark_obj.size[0]) * watermark_obj.size[1])
    
    rsvg = subprocess.Popen(
        ['rsvg-convert', '-a', '-h', str(watermark_height), watermark], 
        stdout=subprocess.PIPE
    )
    watermark_obj = WandImage(blob=rsvg.stdout.read())
    del rsvg
    
    watermark_obj.transparentize(transparency)
    source_obj.composite_channel(
        'all_channels', watermark_obj, blend_mode, 
        source_obj.size[0] / 2 - watermark_obj.size[0] / 2, 
        source_obj.size[1] / 2 - watermark_obj.size[1] / 2
    )
    
    source_obj.save(filename=target)
    
    return True

Minimal usage:

watermark('photo.jpg', 'photo+watermark.jpg', 'watermark.svg')

Why do you use librsvg? Because Imagemagick (which we use via Wand) isn’t as flexible when it comes to scaling vectors. When you ask Imagemagick to scale an SVG, it converts it to a raster PNG of arbitrary size and then scales the PNG, often making it very blurry.

Why don’t you use some librsvg binding? Because it works fine from subprocess and I’m lazy.

What other blend modes are available? See Wand docs under wand.image.COMPOSITE_OPERATORS.

Thanks to Alexander Norov who has helped me assemble this code.
License: MIT.