In my large lecture courses, I often introduce musical examples by displaying a portrait of the composer in question. For composers who lived up through the early nineteenth century, I show paintings. For the more recent composers, I show photographs.
The oldest of these photographs are (unavoidably) grainy, black-and-white affairs. They are an excellent means to illustrate the contemporaneous state of photographic technology, but less than ideal as a means to bring their subjects to life for the students.
Given the recent hubbub over the use of artificial intelligence (AI) to digitally upscale and then later colorize the famous Arrival of a Train at La Ciotat, filmed by the Lumière brothers in 1895, I thought I would see what AI could do for my old photos of famous composers.
Starting with the cropped photograph of Hector Berlioz that appears at the top of this post, taken in 1863 by Pierre Petit, I ran the file through a sequence of three software processes to produce the image below.
First, I upscaled the image by asking the AI at LetsEnhance.io to add missing details to the original file, resulting in an image that was noticeably sharper. Then, I used the image colorization API at deepai.org to add some plausible coloration. Finally, I tapped the “auto adjust” function in the native iOS Photos app to give the exposure a bit more balance.
The resulting color portrait of Hector Berlioz is… not bad. His coat boasts an unusual mix of reddish and greenish hues, but his face is seemingly realistic. With a bit more sleuthing, I likely could have located a higher-quality scan of the original photograph. Even so, the software allowed me to put in so little effort (click, click, click) that I’m personally rather pleased with the results.
What really needs to happen, of course, is for the AI wizards to develop a means for us to hear early sound recordings —such as Brahms playing his own Hungarian Dances at the piano in 1889 —in anything close to the glory of the original performances.