Here’s how Someone Restored Old Photos with ChatGPT

The internet is seeing a new trend in AI images that might breathe fresh life into old family photos and dusty shoeboxes. Users of ChatGPT have found that the AI assistant may imitate an antique, restored, coloured photo.

Eric Hal Schwartz, a freelance writer with TechRadar originally came across this on a Reddit post that, understandably, was initially dubious about ChatGPT’s capacity for repair.

It was much more likely that the AI would “restore” an old headshot by creating a new one of someone who might have been the original photo’s cousin. He put together a prompt that does a decent job at it, thanks to advice from Reddit and Instagram users on how to make it function properly.

“Please upscale and colourise this photo while keeping it faithful to the time period,” is the prompt he ultimately chose. Don’t alter the background, expressions, clothing, or arrangement; just use period-appropriate colour and details. The updated image should precisely cover the old one.

You may consider it a reproduction by a modernist art forager, but it’s not really recovering the image. Better resolution, color results, damage repair, and even re-imaging of certain lost elements are all included.

It’s crucial to keep in mind that this isn’t the same as improving the real image, but it’s funny and feels less problematic than copying Studio Ghibli’s aesthetic.

Eric used some public domain images from the Library of Congress, which is a very great resource, to test the pattern. Below are a few of the findings.

President Theodore Roosevelt will be discussed first. Teddy is shown in the picture working at his desk, writing. A sizable plaque and a vintage liquor decanter with glasses are on the desk. You can see some fine features and colors once ChatGPT has finished its task.

Now, not every detail is accurate. Hand gestures, the shape of a chair arm, a radio that has been swapped out for what appears to be a huge candle, and much more. However, it did a very good job of capturing a lot of the man himself.

The bicycle delivery boy arrived next. Once more, the individual remarkably resembled the original. With pools on a visibly dirty pavement, the original image was already vibrant. For some reason, the colorized version assumed that meant the roadway had a lovely wooden floor.

Nevertheless, the throng in the backdrop and the sign on the structure behind the bike stand out. It even succeeded in capturing the bicycle’s ground reflection.

This soda jerk is going a little too far forward in time. A cheerful man grinning and presenting what he can only presume is a root beer or egg cream float with the style of a Broadway performer portraying a “guy who loves seltzer.”

In addition to the bow tie, ChatGPT’s repaired version perfectly captured his face and emotions. He seems to enjoy his work in the black hole behind the bar, even though the hair is a little strange and he is not sure how pleasant the current drink looks.

Then the two officers with their big caps and long coats arrived. The significant mustaches were probably a requirement for the uniform. Their outfits came to life in the colorized version, which included brass buttons that almost shone and deep navy blue hues.

It’s a picture that goes in the same law enforcement pictures, but the policemen are a touch taller and have slightly better tailoring than their black and white colleagues.

Speaking of moustaches, the last picture’s true star is that one. The person behind it is nearly invisible. It’s architecture, not simply facial hair. This was the most remarkable outcome, with the restored version just marginally removing the stray hairs.

There’s the coat, the face, the haircut, even the shirt crease. I’d want to observe the other members of the barbershop quartet wherever he goes, and every bristle is sharp.

Thanks to ChatGPT for making this possible.