Creating a Photo Mosaic with a Python Script

The Ravenel Mosaic

Objective

Create a mosaic poster using photos taken of the Arthur Ravenel, Jr. Bridge in Charleston, South Carolina. The project started as an attempt to have AI assist with Adobe automations. But it became clear that the more direct and economical approach was to have AI assist in writing a Python script to automate the tasks and mosaic creation without the need for subscription-based photo/graphics software.

The Solution

A Python-based tool that creates a photo mosaic from a collection of images. The tool includes functionality for color profile standardization and intelligent tile placement with variety controls to ensure the same photo isn’t used for entire sections of tiles.

AI Tools Used

While I queried ChatGPT 4o and Gemini 2.0 Flash, I used Claude 4.5 Sonnet out of a sense of familiarity. Final scripts to be added to a GitHub repository (documentation prepared in coordination with Claude).

Features

Size and color profile standardization across all input images
Spectral color analysis and sorting
Intelligent tile placement with variety controls
High-resolution output suitable for printing
Progress tracking during mosaic creation
Detailed usage statistics

Requirements

Python 3.8+
PIL (Pillow) library
NumPy
tqdm

Photo Mosaics with Python

Avoid complicated user experiences by creating your own scripts for prepping and using photos to create mosaics.

Objective: make new things from old photos

I wanted to see if I could create a mosaic from photos I've taken of the Ravenel bridge here in Charleston.

Standardization Made Easy

The script preps the images for mosaic use by standardizing sizes and dimensions, then orders them from 1-149 based on where they fall on the rainbow spectrum.

Creating the Mosaic

The script creates a full mosaic with more than 10,000 tiles placed to approximate the appearance of a reference photo. several refinement tests take about a minute apiece. (See sidebar for tech details.)

The Final Product

Once the trial-and-error is complete, final file generation, of a print-ready image, takes a matter of seconds. The results are prepped for github for others to use.

See the full mosaic