Antietam AI Colorization

[Andy Cardinal]

I was inspired vy @Gettysburg Greg to try my hand at an AI colorization of an Alexander Gardner picture. This photo of Colonel Strong's dead horse has always fascinated/haunted me.

Disclaimer: I am not a proponent of AI, nor do I know what I am doing with it.

General Alpheus Williams: "The number of dead horses was high. They lay, like the men, in all attitudes. One beautiful milk-white animal had died in so graceful a position that I wished for its photograph. Its legs were doubled under and its arched neck gracefully turned to one side, as if looking back to the ball-hole in its side. Until you got to it, it was hard to believe the horse was dead."

 

[Nathan Stuart]

That colorization articulates background details (e.g. grazing livestock, fence posts) not seen previously (by me) in the b/w photograph.

 

[gary]

Makes a sad picture sadder. Virginia Historical Society has a monument to the horses and mules who served during the war.

 

[rpkennedy]

Something that I learned recently is how the AotP was able to recover some of its wounded horses. The units that were looking for their horses would play the feed call and the animals that were capable of doing so would stagger towards the sounds of the bugles. It makes one realize that they were every bit the soldiers that the humans were.

Ryan

 

[Zack]

Thank you for posting!

The AI did a pretty good job. It did seem to have some trouble with a few aspects of the photos.

The original can be seen here: https://www.loc.gov/pictures/resource/ppmsca.34150/

I think in a few cases the AI interpreted marks on the photograph as part of the image. For example:

The AI seems to have interpreted this as a mushroom. To my eye, though, it's a scratch.

You can also see the AI had some trouble with that horse in the background. It gives the general suggestion of a horse, but it's a bit wonky if you look at it too long.

The horse at the far left seems to have a saddle or possibly even a rider on it.

The middle horse in the AI photograph I don't see in the original, and the one at the right I can't make out in the original clearly.

It also seems to have generated a barbed wire fence based on the vague background...vertical item...I'm not sure what it is.

I also can't really tell if the figures on either side are people or foliage and the AI seems to have split the difference in some cases.

I do like seeing all the green - it's easy to forget when looking at these pictures sometimes that the ground isn't just mud and dirt - although the grass seems a bit more trampled in the original photograph than the AI shows.

I don't point these out to critique the OP - it's just interesting to see what the AI made out of some of the more difficult elements of the photograph.

It is striking though how much more it hits in color!

 

[Andy Cardinal]

Thank you for posting!

The AI did a pretty good job. It did seem to have some trouble with a few aspects of the photos.

The original can be seen here: https://www.loc.gov/pictures/resource/ppmsca.34150/

I think in a few cases the AI interpreted marks on the photograph as part of the image. For example:
View attachment 582152

The AI seems to have interpreted this as a mushroom. To my eye, though, it's a scratch.
View attachment 582153

You can also see the AI had some trouble with that horse in the background. It gives the general suggestion of a horse, but it's a bit wonky if you look at it too long.
View attachment 582154
The horse at the far left seems to have a saddle or possibly even a rider on it.

The middle horse in the AI photograph I don't see in the original, and the one at the right I can't make out in the original clearly.

View attachment 582155
It also seems to have generated a barbed wire fence based on the vague background...vertical item...I'm not sure what it is.

I also can't really tell if the figures on either side are people or foliage and the AI seems to have split the difference in some cases.

I do like seeing all the green - it's easy to forget when looking at these pictures sometimes that the ground isn't just mud and dirt - although the grass seems a bit more trampled in the original photograph than the AI shows.

I don't point these out to critique the OP - it's just interesting to see what the AI made out of some of the more difficult elements of the photograph.

It is striking though how much more it hits in color!

Thanks for the analysis. I noted what appears to be the barbed wire fence but not the others you pointed out. I imagine if I knew nore about using AI I might be able and go back and fix those things. This was my first attempt at using AI for mich of anything. Truthfully, I am not sure what ai think about it, but like you I like the color, and I also like that it reveals some things that I hadn't really noticed before in the picture.