Bringing New Life To A Century-Old Photo Through AI

By Published On: January 19, 2024Categories: LearningTags: ,

This original photo captured a special moment over 100 years ago in 1921. Meet Jack Bunch (who I was named after). Roommates with Glen Miller back in college, my grandfather was a prolific behind the scenes guy during the swing era, but was also in the first Ripleys Believe it or Not for playing 3 clarinets at once, and went on to be the head of music at ABC studios for 15 years. Back then, the technical limitations of cameras produced still images frozen in time. But by leveraging modern AI techniques, we can now digitally transport this decades-old snapshot into an extrapolated motion picture revealing new perspectives.

Expanding Spatial Context With Image Extrapolation

The first step involved taking the vintage photo and using AI to conduct something called  “image extrapolation” to predict and fill in new visual information beyond the original frame. Although I have done this hundreds of times in Photoshop, this is the first time I attempted it in Canva. This expanded the spatial bounds to reveal more of the environment and scene context while inferring realistic textures and details based on the source content.

Once I brought the smaller image, into a much larger design size canvas, I then selected the photo, then photo edit, then magic expand. Like Photoshop, it actually gave me 4 options to choose from and I have to say, I was blown away at the quality. The expanded context felt very real and had a lot less anomalies than when I had done this in the past. The result looks like this:

Whereas the 1923 photographer was constrained by technology to a narrow field of view, extrapolation allows us to pull back the lens and envision the broader world the photo captured just a slice of. Details once confined to imagination come to life. People have done amazing experiments like this one. For example take a close shot of half your face, and see what AI extrapolates the rest looks like. Or focus on just your eyes and see what it things you look like beyond the cropped image.

You that I was done didn’t you? Let’s bring motion to the picture

Next, extending beyond just visual expansion, temporal manipulation through AI opened additional possibilities for rediscovery by forecasting how the static scene may have looked as a moving picture. Advanced machine learning techniques studying the image contents can make informed guesses about likely motion vectors and temporal transformations to inject. This can produce convincingly realistic motion and animation. What you see below is an 8 second video that was extrapolated during a subtle pan and zoom. Now of course the detail is riddled with anomalies and strange artifacts, but you can see where this is going. If you ever watched Star Trek and remember the holodeck… this is holodeck Alpha version 1.0.

The output video illustrates the plausible dynamics of the scene over a few seconds – telling a mini story through the details and movement patterns invented by the ML model based on its computer vision insights.

This cross-pollination of AI photo expansion and motion prediction techniques allows us to transcend restrictions from the past. We can rediscover and reimagine historical snapshots like never before thanks to the digital time machine provided by modern computer vision and creative ML tools.

Let me know if you would like any sections expanded or have additional ideas for covering AI photo/video expansion and enhancement! Aiming to give a solid introductory overview of capabilities and techniques.

Share with friends!

want to learn more?

Sign up for our email newsletter or contact us directly