About this project

As we watch the death count rise each day, eventually we become numb to it. Even if we understand intellectually that each of those numbers represents a real person, it is borderline impossible to fully grasp the human toll of covid-19.

A memorial which names the dead in their entirety, like those created after the Vietnam War and 9/11, will never be possible for covid. So many have died alone at home, having never been tested. Government death records provide incomplete data and the results of laboratory testing are rightly anonymous to the public.

How can we mourn the death of people whose faces we'll never see and whose names we'll never learn? How can we comprehend the scale of it?

I hope this website will help visitors to better understand the covid crisis and better process their grief—I know it has done so for me.

Technical Process

The photos on this page were sourced from generated.photos. Some were digitally aged in faceapp to account for the under representation of elderly people in that service. I first encountered the concept of generative media through This Person Does Not Exist, which was built to showcase the StyleGAN2 library. I assume the photo generation service I used leverages StyleGAN2 under the hood, but I do not know this for sure.

The age and gender breakdown of covid victims was sourced here. The racial breakdown was sourced here and here. These statistics are admittedly imperfect due to gaps in reporting. For instance, there is no data breaking down victims by race and age simultaneously. While we know with some confidence that roughly half of covid victims have been over 75 years old, and one-quarter of covid victims have been African American, we don't know for sure that one half of African American victims have been over 75.

It may be that the victims within some demographic groups skewed older or younger than the national average. It also may be that some demographic groups are uniquely likely to go unreported. I have used only the data recorded in official sources, which most agree is an undercount of the actual total. If more accurate or complete data becomes available, I will update this page.

That said, I feel the use of computer generated images helps to smooth out some of the imprecision inherent in this project. The faces shown are not real, and the people shown have no true age or race. So, while it's true that about half of these images appear over 75 years old to my eye, that's merely a subjective assessment. There is no way to objectively state that the images I have selected are too old or too young, because the photo subjects have no objective age. I have also embraced racially ambiguous images because such ambiguity exists in real life.

This page uses 300 unique images, each one repeated in a random order enough times to equal 535,000. I originally planned to use one unique image per covid death, but this proved unworkable due to the massive file size entailed in hosting so many pictures. Even with very small images, the entire page would have weighed several gigabytes, and likely consumed many terabytes of server bandwidth. This would have pushed hosting costs into unattainable levels, making the project impossible. I decided to use 300 images after some experimentation, because with any fewer images, the repetition starts to become too noticeable.

The total collection of images is a microcosm of the actual victim pool. For instance, 15.7% of reported deaths have been people of Latin American ancestry, which equals 47 out of 300 images. Of these, 61% have been men. So 29 of the 47 Latin American images are male. By picking a random image out of a representative pool 535,000 times, we achieve an overall list roughly equivalent to the real world. For groups that constitute fewer than 1 in 300 deaths, I added special handling in code to ensure representation.

I briefly considered collaging real-life names or images of covid victims, but I ultimately concluded that this would not be appropriate without the explicit approval of the victims' families. Perhaps one day a fitting memorial to the victims of covid will be constructed based on voluntary participation of the bereaved, akin to the national memorial for victims of HIV/AIDS.

When I began this project, it was called "100,000 faces." I never imagined that the death toll would rise to more than 5x that number. For a time, I kept updating it. First at 150K, then 200K, then 300K. Eventually, it just became too painful to keep watching the death toll rise, and rise, and rise, and I abandoned the project for a time. After a year of death and grief, and I decided that project would no longer be updated, and it would remain frozen in time as a memorial to those we lost in year 1.

Similar Projects

There have been other efforts to create virtual memorials to the victims of covid-19. Of these, the most eloquent and moving is surely “An Incalculable Loss” by the New York Times. The New York Times piece came out the day before I initially planned to launch this project. It is such a moving and complete tribute, that I considered scrapping this project entirely, because it is hard to imagine a more effective memorial than the one already published. After some delay, I decided there is still some value in both of these projects co-existing because human faces have a special, unmatched emotional resonance. Several other news sources, including ABC News and the Washington Post have endeavored to collect real-life images of known victims. NPR has collected a list of other memorial projects, both virtual and in the physical world.

About the Author

This project was created by Matt Korostoff and is not affiliated with any organization.