Colorizing History: How I Used ChatGPT to Breathe Life into My Ancestor’s 1903 Obituary Photo

File under: science fair project for adults.

But like, a fun one. No styrofoam planets. No baking soda volcanoes.

Not only am I a genealogist, but I also consult with business owners to make their businesses run smoother than a sea otter on a slip-n-slide.

So, when AI started changing the way business owners work, I dove headfirst into it.

I also began using it almost immediately in my genealogy research. There are many ways to utilize AI in your research. I’ve attended some AI for genealogy workshops that've been great, and just as many that've really lacked.

When it comes to colorizing old photos, I’ve developed a method that I think works pretty well.

My 2nd great-grandfather was a mystery in my family tree for decades. Born in Vilna, Lithuania, he disappeared from the records sometime around 1865, shortly after my great-grandfather's birth. No death certificate. No immigration trail. Nothing in the Lithuanian State Archives. Just a gap in the tree and me wondering: What happened to him?

Until... boom. An obituary from New York in 1903. With a PHOTO. (I’ll have to do another blog post on how I found his obituary!)

I wanted to colorize it immediately and look into his face. I wanted a version that honored what he truly looked like as much as possible, that didn’t distort his face into an AI’s best guess, and could give me a peek at what he might have looked like in full living color.

So I pulled out my prompt-crafting tools, called in ChatGPT, and got to work.

Before and After of an image from around 1890 of my 2nd great-grandfather using Chat GPT to colorize the image

Why Use ChatGPT for Genealogy Colorization?

There are loads of resources to colorize old photos. I am not saying that Chat GPT is the best one, it’s just a platform that I work in more often than the others, and have seen good results from it.

However! There is a huge catch with all AI-generated photos:

It fills in the image gaps with its best guess.

It doesn’t know what color your ancestor’s eyes were (unless you tell it) or how sun-worn their skin looked from 30 years of rooftop chimney work, or in my case, pawn shop ownership.

That’s where ChatGPT, combined with a solid prompt and a reference image, comes into play.

The Step-by-Step: How I Did It (And How You Can Too)

1. Log into ChatGPT

Visit chat.openai.com

2. Prep your image(s)

Start with one clear black and white or sepia image of your ancestor. This works best with a simple photo, featuring minimal background noise, and ideally without a ten-person family reunion in a flowery field. One face, center stage, basic background.

💡 Pro tip: Keep a reference image on deck, a clearer, more recent photo of a relative who resembles the ancestor (like a son, daughter, or sibling). You’ll only upload this if your first result comes back looking... off.

3. Write Your Prompt (with purpose)

AI consumes a significant amount of energy with each prompt and image output. You want to include as much detail as possible in the original prompt to minimize your interaction with it. I like the idea of green genealogy, and I’m not talking about graveyards!

Include everything you know about the person and the photo:

• Date or range the photo was taken

• Location (city/state/country)

• Their birth year and birthplace

• Religion, ethnicity, or cultural details

• Occupation, if it helps (e.g. “coal miner” paints a different picture than “concert celloist”)

• Known features (like eye color from a passport application)

Important: Ask ChatGPT not to add extra details and to preserve the facial structure.

“Prompt:

This is a photo from a 1903 New York newspaper obituary of an Orthodox Jewish man. This man was born in Vilna, Lithuania around 1830 and immigrated to New York around 1875. The photo was taken between 1885–1900 in New York.

Do not add any extra details.

I understand that automated colorization tools may unintentionally soften, sharpen, or adjust facial details, especially with older, low-resolution images. Please preserve the original facial structure as closely as possible.”

4. Upload the image

In the same message box as your prompt, click the small ➕ icon in the bottom-left corner to upload your old photo. Then hit send.

5. Wait (Patiently-ish)

It takes a while for AI to do its thing with images, hence the drain on the environment and why we want to keep its use to a minimum. Use this time to take a nice deep breath with a slow exhale (maybe twice.)

6. Review the results and get into conversation with ChatGPT

Did your ancestor come back looking a little… off? Or like they’re starring in a Tim Burton film? Time to fine-tune.

Tell ChatGPT what worked and what didn’t. Be specific. You can even ask it:

• “What assumptions did you make?”

• “What parts of this face did you adjust?”

• “Can you ask me questions to get this closer to the original?”

If the face still isn’t quite right, this is when you bring in that reference image.

Upload the photo of the family member who resembles the ancestor, and use a new prompt like this:

“This is the same request as before, but I’m now including a photo of his son. I feel they look alike and you can use the son’s photo for facial reference if needed.

Do NOT merge the two images. Do NOT apply the son’s features directly. Only use the second image as a guide to help preserve the facial structure.”

What Makes This Approach Different?

Most people upload an old photo into an AI and give the prompt “colorize this image”. That’s like asking a stranger to decorate your home without telling them you hate wallpaper.

The prompt is missing:

• Context

• A reference image

• The details that will limit your interaction with AI, reducing conversation time.

Before You Upload and Share the Colorized Photo…

If you plan to add your newly colorized image to Ancestry, MyHeritage, or wherever your family tree lives online, disclose that it was generated with AI in the photo details. These colorized photos are artistic interpretations, not photographic facts. It’s tempting to treat them like historical evidence (because wow, they look real), but unless you time-traveled with a Kodak into 1903, they’re still best guesses.

A beautiful, compelling guess. But a guess nonetheless.

You’ll also want to include the original photo, so that anyone viewing the images can see the difference between the two.

Pass It On

Do you know someone with a box of forgotten photos? A genealogy buddy who could use this tip?

Send them this blog.

Try It Yourself and Tag Me!

Got an old photo burning a hole in your hard drive? Colorize it with ChatGPT.

Then post your before-and-after to social media and tag @ourchives.co so I can cheer you on!