On a side note, I find it quite curious how the resolution on this image from WAttention [1] mismatches. From the OP, the couple portrait attributed to Wikipedia, which is the latest 2018 retouched version [2], has a much higher resolution and specifies @Press as the source. The 2016 original version [3] matches the couple portrait on @Press [4], which has the same blueish hue and the additional artifacts. This links back to Nakamuraya's website itself [5], which has the same food photo combination as [1], but in consistent resolution.
At first, it seems to me that the couple portrait on [1] was (intentionally?) degraded because of the different lighting than [5], and the food photo is simply another version that was available during the retrieval (notice the crop). On second thought, it could have simply been another version that was available from Nakamuraya's. TIL images uploaded to Wikipedia are encouraged to be enhanced [6].
I wonder, as AI tools getting more pervasive, how many variations with artificial details will exist due to constant upscaling/downsizing? (Accurate replication might not be important in certain contexts, but I imagine artificial details will eventually leak into the chain of reference)
Of course, old photos like this are scanned/photographed into digital format in the first place.
[1]: https://wattention.com/wp-content/uploads/2015/07/Shinjuku-N...
[2]: https://upload.wikimedia.org/wikipedia/commons/f/f5/Rash_Beh...
[3]: https://upload.wikimedia.org/wikipedia/commons/archive/f/f5/...
[4]: https://www.atpress.ne.jp/releases/104283/img_104283_8.jpg
[5]: https://www.nakamuraya.co.jp/common/img/top/pavilion_img07.p...
[6]: https://en.wikipedia.org/wiki/Wikipedia:How_to_improve_image...