GEMINI - Behind the Darkroom: How We Restore the Archive

When you look at the restored photographs across this archive, you aren't looking at basic color filters or manual Photoshop painting. Instead, every image passes through a custom-engineered digital restoration pipeline powered by generative artificial intelligence—specifically, Google's advanced multi-modal vision engine, nicknamed "Nano Banana" in our workshop.

But how does a modern cloud engine look at a black-and-white photo from 1960 and "guess" the exact colors of a school jersey, a muddy cross-country field, or a brick pavilion?

The short answer is: It doesn't guess. It analyzes, adapts, and infers based on historical data. Here is how the magic happens under the hood.

1. The Language of Light (Semantic Attention)

The AI doesn't see a photograph the way we do; it breaks the image down into a dense mathematical map of shapes, textures, and contexts, a process driven by what computer scientists call spatial attention mechanisms.

When a black-and-white photo is fed into the pipeline, the engine analyzes the grayscale values (the "luminance") and compares them against billions of real-world color data points it has studied. It recognizes that a specific texture is "wool fabric," another is "brickwork," and another is "muddy grass." By understanding what the object is, it can infer what color it should be to look historically realistic—applying warm, natural skin tones or separating the exact texture of old masonry from the background.

2. Guarding Historical Truth (The Context Anchors)

Left to its own devices, an AI might try to be "creative" and invent color schemes that never existed. To keep the archive completely accurate, our pipeline uses a technique called Few-Shot Multimodal Prompting.

Before the AI even looks at a vintage school photo, we feed it Reference Anchors—such as a modern color photograph of an original school uniform, a physical tracking card of school ties, or an official blazer badge.

We hardcode our pipeline with precise, direct instructions:

"Refer to Item #3 in the reference image. Match the purple body and gold sleeves for the hockey shirts."

Because the AI reads these reference pixels first, its attention mechanism locks onto those exact color values. When it processes the black-and-white team photo immediately afterward, it copies and applies those precise purple and gold hues directly onto the old schoolboy jerseys, ensuring 100% historical accuracy.

3. Locking Down the Geometry

One of the biggest risks with generative AI is "hallucination"—where the computer tries to redraw a blurred face or alter an expression. To prevent this, our custom Python script forces the engine to run at a very low "temperature" constraint.

In AI terms, a high temperature allows the model to be artistic and take wild risks. By locking our pipeline down to a rigid, icy-cold temperature, we strip away its creativity. We force it to act strictly as a preservation tool. It is ordered to respect 100% of the original film grain, face shapes, and background structures. The AI is only allowed to restore clarity and paint the scene—never to reinvent it.

4. Running a Professional Enterprise Pipeline

To handle thousands of images without breaking a sweat, our backend workflow operates exactly like a professional software development environment:

• Sleek Automation: We utilize a custom Python script that automatically handles image scaling, formats the data payloads, and communicates directly with cloud servers.

• The Cloud Backup: The entire codebase, custom configurations, and development history are managed using Git and synchronized daily to a secure cloud repository on GitHub, ensuring this historical project is permanently backed up and protected against local hardware failures.

• Overcoming Limits: To circumvent strict structural boundaries—like platform image limits—our architecture target-scans and prioritizes core database preservation pages, ensuring the archive remains incredibly fast and permanently accessible to the community.

A Labor of Love

Technology provides the engine, but the real heartbeat of this website is the community. Every time an old boy emails in to identify a missing name from a 1972 athletics photo, or we publish a new gallery to our popular "Informal Drinks" section, we are proving that the bonds formed here decades ago are still active and vibrant today.

We aren't just colorizing old celluloid; we are keeping our collective memory alive.

Tips for Customizing This Page:

• Add Your Personal Touch: Right at the beginning or end, you can add a paragraph about how you spent weeks fine-tuning the code workbench in VS Code, or mention how much joy it brings you to see an old, "appalling" photo finally pop into pristine clarity.

• Visual Appeal: If you have room on the page, putting a Before & After slider or side-by-side image of that uncropped 1972 junior athletics photo (showing the raw B&W vs. your gorgeous purple/gold finished output) right next to this text would be the ultimate visual proof of everything described above!

CLAUDE - Bringing the Past into Colour

How artificial intelligence is transforming our archive — one photograph at a time

The Challenge

The photographs that survive from St. Nicholas Grammar School's years span two decades of school life — classrooms, sports days, concerts, staff portraits, and the everyday moments that make a school community what it is. But time has not been kind to many of them. Printed on the photographic paper of the 1950s, 60s, and 70s, many images have faded, yellowed, or deteriorated over the decades. Even the sharpest originals were, of course, captured in black and white — the standard of the era.

For years, these photographs told their stories in shades of grey. They were historical, certainly — but somehow distant. Harder to feel.

Why Colour?

There is something remarkable that happens when colour is introduced to an old photograph. A grey playground becomes a real place. Grey faces become real people. The psychological distance that black and white quietly creates simply dissolves.

This is not about altering history. It is about connection. When a former pupil sees a colourised image of their old school for the first time, they are not seeing a document — they are seeing a memory. And for younger family members who never knew the school, colour makes it vivid and immediate in a way that monochrome simply cannot.

That was the thinking behind our decision to begin colourising the archive. Not to replace the originals, but to offer something alongside them — a new way in for people who might otherwise scroll past a faded grey image without pausing.

How It Works

The colourisation is carried out using Nano Banana AI, a sophisticated artificial intelligence tool trained on vast libraries of photographic reference material. Rather than simply applying a colour filter, the AI analyses each image in detail — identifying objects, textures, and context — and then makes informed decisions about likely colours based on what it has learned.

Grass is rendered green. Skies are given appropriate blues. Skin tones are carefully estimated. School buildings, sports kits, and everyday clothing are all interpreted as intelligently as the available visual information allows.

The results are, in many cases, quite remarkable.

Our Workflow

Colourisation at St. Nicholas is not a one-click process. Each image passes through several careful stages before it appears on the site.

Original photographs are first scanned or sourced digitally, then prepared — correcting exposure, reducing damage, and optimising contrast to give the AI the best possible starting point. The image is then processed through our colourisation pipeline, which applies the AI's interpretation consistently and at high resolution. Finally, each colourised result is reviewed and refined manually using professional photo-editing software, adjusting tones, brightness, and colour balance until we are satisfied that the result does justice to the original.

It is a labour of love — but one we feel the archive, and the people in it, deserve.

Before & After

[A selection of before-and-after image comparisons will appear here — showing the original black-and-white photograph alongside its colourised version.]

We think the results speak for themselves.

Honest Limitations

We want to be transparent about what AI colourisation is — and what it is not.

The colours you see are intelligent interpretations, not verified historical fact. We cannot know with certainty what colour a particular pupil's jumper was, or the exact shade of paintwork on a classroom wall. Skin tones are estimated from context. Uniform colours are based on what is known about the school, combined with the AI's broader knowledge — but they may not be exact in every case.

Where we have reliable information about specific colours — from written records, contributor memories, or surviving coloured materials — we use it. But we ask visitors to enjoy these images in the spirit in which they are offered: as a thoughtful, carefully produced interpretation, designed to bring the past closer, rather than as a photographic record of fact.

The original black-and-white photographs remain the definitive historical archive. They always will.

A Living Project

Colourisation of the St. Nicholas archive is ongoing. New images are being processed regularly, and we are always looking for additional original photographs to work with.

If you have old photographs from your school days — or from a family member who attended St. Nicholas — we would love to hear from you. The better the quality of the original, the more the AI can do with it. But we will always do our best, whatever the starting point.

Every image recovered, restored, and brought back to life is one more piece of a shared history that might otherwise have been lost.

For a note on the interpretive nature of colourisation, and how to request the removal of any image, please see our Disclaimer.

Happy to adjust the tone, length, or any section — and once you're happy with it I could look at turning it into a proper downloadable page file if that would be useful.