The first Pokémon movie had a bit of an odd production history, and you may have noticed some shots are sharper than others. There are some technical reasons for that, which I explain. I also provide some other options, courtesy of AI image enhancement.
Technical Details:
The movie was originally produced on 16 mm film, as opposed to the 35 mm film normally used for theatrical movies. Around the time the movie was being brought to the west, some shots were re-done in 35 mm, hence the difference in visuals.
I wanted to see if it was possible to make the 16 mm footage look sharper. In earlier times, this could be done by using a sharpening filter on the image, but this would cause haloing artifacts that would make the sharpness look fake. With the advent of image processing AI, I decided to use a deblurring model I had created using the ESRGAN framework. I experimented with blending the model with a base untrained model to get the results you see here. Overall, I'm fairly pleased with results in the 40-50% strength range. Values above that tend to bring out too much of the film grain, making it look as if the film had a higher ISO, and creating a haloing effect.
Looking beyond what I've done here, it should be possible to create a more accurate model by creating high/low-resolution pairs of footage on 35 and 16 mm film (likely a 16 mm print created from a 35 mm master, or perhaps specially created footage for such a model). This would likely avoid the grain issue mentioned above.
