While the update was generally well received, there were a few who reacted with skepticism. So I took some more tests to demonstrate the repeatability of the phenomenon.
1. Take a 30mp test photo.
2. Record a 5 minute 4k video.
3. Take a second 30mp test photo.
4. Compare the two test photos.
I followed exactly this procedure, using nothing but the camera (with the self-timer). I positioned one lens very near a wall, so that we could see if the blur issue affected both near and far objects. Based on Tamas Pasztor's explanation (which I agree with), the sensor or processor heats up and changes the focal distance from hyperfocal / panfocal to a macro-like distance.
To rule out stitching or any postprocessing problems, I just compared the unstitched photos.
Test shot 1 (ISO 100, f/2, 1/2000)
Test shot 2 (ISO 100, f/2, 1/2000)
The results are the same as before, and confirm Tamas Pasztor's findings.
This is a comparison of 1:1 crops at a farther distance. The left one is Test shot 1, the right one is Test shot 2. You can see that despite the identical exposures, Test shot 2 is more detailed.
At the same time, if you look at a close object (the wall), Test 2 is not blurred - in fact, it has slightly better acuity and contrast.
To summarize, these are the observations:
- When the camera continues to operate for some time, later photos become blurry compared to earlier photos.
- The farther the object is from the camera, the greater the blur.
These results are perfectly explained by Tamas' findings, i.e. the camera heats up, there's a part inside that expands from the heat, which moves the lens slightly, and changes the focal distance.
Here are copies of the actual test shots so you can pixel peep as much as you want: