One of the things I’m coming to appreciate with AI renders is how easily it can make the smallest of changes.

This is just a colorized version of a favorite bit of CG silliness from a few years back. I didn’t do, um, *either* of the originals. I just had GPT Image colorize it.

Years ago, someone had started playing around with this picture. We have a US Navy photo from 1942. A Brewster F2A-3 Buffalo has crashed on the deck of the USS Long Island. The Long Island was an escort carrier that spent the War doing training (carrier qualifications) and aircraft deliveries (unlike a cargo ship, an aircraft carrier can deliver planes fully assembled). The date is easy to confirm visually, the colors and markings on the Buffalo were in use from June 1942 to February 1943.

I love this image. It was made in 2010 or earlier. I had it up on my locker at work for years. It is funny, the AI took some liberties I’m not thrilled with. Like it introduced an error to the X-Wing, notice how on the “original” here we have a straight line from the top of canopy to the nose cone. On the re-render the AI adds a notch where the canopy joins the forward fuselage.
Also the men in the image all look like War time USN crewmen. While in the color version they look more civilian. I might have been able to fix that with a better description of how the people were dressed, at least locking down colors?
I don’t know what I could have done about the error in the X-Wing’s shape. I did two renders side by side, the other was more wrong. It changed the framing details on the canopy and clearly made one of the deck crew a woman. Seriously! Everyone knows the USN did not have any women assigned to X-Wing handling in 1942! I’m pretty sure that can be proven…

2 responses to “More Goofing Around…”

  1. Zeno Avatar

    All things considered it really did do a great job colorizing everything. What did you have to give at besides the source image? An x-wing picture?

    I’ve been amazed lately in several of my renders – like in the elf village or the wheel clan mines. I fed it isometric screenshots from the game and asked it to use them to generate side-view imagery of characters doing things in those areas. And it pulled it off! Believably and recognizeably! Same with the ancient submarine on the Isle of Despair. Changing the view angle by 90 degrees really demands a good understanding of the three dimensionality of the image. I was shocked how well that worked! You ought to try it with Icewind Dale screenshots sometime, if you haven’t already.

    It really is difficult to wrap your head around how much of our conception of imagery is changing with this new technology. When everyone can easily manipulate photos in seconds in ways that used to require special effects experts – with less obvious artifacts… Its just insane. A photo isn’t proof of anything anymore. It won’t be very long at all before the same is true of video. And then what happens to court evidence when any two-bit lawyer can discredit surveillance video and the like? Political Ads when you can easily depict anybody saying or doing anything? Suddenly they can get away with the old “say the completely opposite thing to two different audiences” trick again because they can tell each group the other video is AI. Wild times.

    Like

    1. atcDave Avatar
      atcDave

      I think we’re well into the period where unverified photos prove nothing in court. I’m sure it has to be from some sort of secured security camera, or multiple witness photos showing the same thing. It is clearly to easy to manipulate. Photos now should follow “chain of custody” type protocols like any other evidense.

      This one, I didn’t do the original fake. I found it on line many years ago. At the time, I was awed by how well done it was. It still represents two very favorite interests of mine, so I guess its become meaningful to me. So I just colorized it, and am both impressed with how easy it was. And annoyed by some minor errors that croped up. I think the AI got bored and was messing with me!

      Yeah I’ve used screen caps quite a lot for current renders. Especially back grounds. Take a screen cap of the isometric background and tell the AI to change angles, resolution, lighting. It is all amazing. I’m working on the last one now and am using screencaps and art from D&D sourcebooks. And layering things in to avoid getting too much going on in individual scene descriptions.
      I’ve also started working on my next BG run, going back to a photographic style. I can use most of the same techniques, but it is fussier in some ways because the viewer is more likely to pick on things that don’t look right. Trying get some depth of field things to look right, all in all its very impressive what it can do. But as we’ve commented before, as it can do more we want to do more. And that is never completely easy.

      Like

Leave a comment