May 2025.
Experiment in Neural Filters.

I recently discovered a filter tool in Photoshop that helps restore old photographs. I thought I’d see how it works. I found a scan of an old printed snapshot that I mistakenly scanned at 475×330 pixels, taken in 2001. To start with the restoration, the first thing I needed to do was upscale the photo. I’ve always had good luck with Luminar Neo’s Upscaler, so I up-scaled the image by 2X (950×660).

(Johnbo from the future here: After completing this project, I discovered that Photoshop has an upscaler built in, so I could have done all the pre-processing in Photoshop. I’m not a regular Photoshop user, so I didn’t think to see how its upscaler works.)

I could have upscaled the photo by 4X or more, but for an Internet web page, that would only increase download time and not provide any real improvement. I also thought that it may create more negative artifacts.

I loaded the upscaled image into Photoshop and used the Photo Restoration Neural Filter. It’s in the Filters menu under the Neural Filters submenu in Photoshop 2025. It did a nice job fixing facial features, but I discovered that you don’t want to zoom in too closely, as the image quickly falls apart.

I then exported Photoshop’s TIF file into Luminar Neo for final processing. I used the Noiseless module to remove noise from the low-light areas and the Supersharp module to sharpen the edges in the photo.

When I returned the photo to Lightroom Classic for cataloging, I noticed the shine from the camera’s flash on Josh’s nose and upper lip. I used a clone stamp to remove the shine and upscaled both images to 1024×711 to use the Compare tool in WordPress for the opening photo.

Then I started on a more ambitious project, two photos that were the worst of the worst. Around 2002 or 2003, a friend and I flew to Chicago to land at Meigs Field. I had these photos, and I knew they were unsalvageable, but a few years ago, I decided to tell the story of why we would go to Chicago to land at a soon-to-be non-existent airport. That story is here if you want to read about what Mayor Daley did to the popular downtown Chicago airport.

In that story, I used a highly cropped image dressed up as best as I could do at the time. For this project, I thought I might try the neural filter to see if I could salvage the unsalvageable. They are still not worth the time it took to “rescue” them, but they are better than they would have been a few years ago. For academic reasons, I’ll show you the originals and the final copies, which were converted to black-and-white because the color restoration was unacceptable.

Meigs Field jutted out into Lake Michigan on Northerly Island. The field was closed in 2003 and became a park and nature preserve. This color image was from a film camera I owned at the time. These were scanned from snapshot prints. It wasn’t foggy, and conditions were not the best, but we had no safety issues. The photos make the weather look worse than it was. You can see there was some blue sky.

I cropped the photo to better view the single runway. I processed it with a neural filter, noise reduction, sharpness, and conversion to black-and-white. The vertical lines in the original color image are much less noticeable in black-and-white, but the neural filter could not remove them.

As we turned to make our final approach, I snapped this photo while my friend was flying the plane. We were only moments from landing. Time wasn’t kind to either of these photographs on paper, and this one needed the most work.

For this photo, after applying the neural filter, I used content-aware fill to remove the plane’s panel and replace it with an AI-generated extension of Lake Michigan. Again, I used Luminar Neo for noise reduction, sharpness, and conversion to black-and-white.

The Restore Old Photographs filter is something you should try if you have photos that need a little extra processing to make them new again, but don’t expect a miracle. Overall, the picture of Josh at his graduation turned out quite well; the other two… not so much.

John Steiner