AI colorization with ChatGpt experiments.

[JSteinman]

 

[moon puppy]

To me, the tank color seems too uniformed. Shadows are not well done, look at the spare wheel rack in the BW photo, very dark in the shade of the side armor of the turret.
I like how the car turned out.

 

[Sharkmouth]

The redeeming feature is that some details not easily seen pop out, like the band at the front of the turret skirt. For that reason, I like playing with it but wouldn't save it into my files unless adding notes like this:

 

[JSteinman]

You guys are correct, but your criticism doesn’t negate the usefulness of the exercise. It allows a better “immersion” into the scene and provides a better feel for me as I model a subject. It’s a tool like any other.

 

[JSteinman]

To me, the tank color seems too uniformed. Shadows are not well done, look at the spare wheel rack in the BW photo, very dark in the shade of the side armor of the turret.
I like how the car turned out.

I guess it is also about the control. The prompt used was as follows:
Archival colorization, push the chroma, no other changes permitted.

 

[Sharkmouth]

You guys are correct, but your criticism doesn’t negate the usefulness of the exercise.

I think it has its merits. I simply won't save it as I have many original WWII color photographs and don't need confusion.

 

[paddy]

I think this sort of thing is great if its for private use but the Internet is awash with badly coloured or wrongly coloured B&W pictures. You often find people arguing over a point of detail like a rivet pattern or a marking and completely ignoring the fact the subject is the wrong colour.
With my current build , tying to find accurate original colours for Desert/Far east RAF schemes has been a nightmare as just when you think you have found the answer it turns out to be another colourized picture.