The damage I'm seeing is consistent with fire damage alone, with only cursory evidence attributable to collision or debris. The car roofs are depressed to a degree consistent with the burning of interior elements, followed by softening of the roof elements and a partial collapse under its own weight. And depending on timing. the evidence could be considered consistent with fires having burned to the point of exhausting their fuel and burning out. Smoldering is hard to detect in still images (especially through
@nt1's wild excursions through the luminosity adjustments).
No, the effect doesn't occur with any image. It's easy to believe
@nt1 attempted his manipulation on photos he took himself and did not see any similar effect because his images were of too high a quality.
Tl;dr, the sharper the image, the less susceptible it is to the anomaly
@nt1 is inducing.
My avatar is currently being oversampled as the forum displays it. The image snippet I submitted years ago was suitable for the old forum software's notion of proper size. But the new forum wants to display larger avatars. The stored image has too few pixels, so the browser is resizing the image at the forum software's request to be larger in terms of pixels. But it can't put in data that was never there. So it punts by oversampling, and that's why it looks blurry.
As I pointed out in my demonstration a few pages back. oversampling is what makes the effect appear. That won't occur in photos you take yourself, unless you zoom, crop, and resize a portion of the image in order to induce oversampling. The images he's using are digital enlargements of details in larger photos. Plus he's obtained them from convenience sources which may have done the zooming and resampling themselves without his knowledge. They all exhibit poor image resolution despite the presence of many pixels, much as does my avatar.
If you have an image with a dark region and a light region as part of a detail, the transition between those elements in a well-sampled, well-focused photograph may be only a very few pixels wide. If you have a 256x256 pixel region from that image and you blow it up to occupy a more viewable 1024x1024 raster, the additional pixels that have to be added get their values by interpolating between pixels in the original image. There are common methods (e.g., linear, bicubic) to do this mathematically. But the result is inevitably a larger region of pixels that transition between the former dark and light regions. Instead of being only a couple of pixels wide as in the original image, that transitional area in the resampled (enlarged) image could be dozens of pixels wide—occupied by these new, smoothly changing intensities. Then his brutal manipulation of the luminance map remaps that smooth transition into a single delineated region that doesn't actually exist in the image data.