My feelings on the project are complicated.
On the one hand, it is clear that he has spent a vast amount of time and money to bring together a large collection of sources for a restoration, and he clearly believes that the results have been worth it. I appreciate and applaud the enthusiasm.
On the other hand, the results he touts in his videos of doubling the detail in the final result simply aren’t apparent to me. The resultant image has less grain, but the way he has described his detail recruitment algorithm makes it clear that there’s not much benefit in stacking sources and the majority of the detail may be coming from temporal super resolution instead. The issue is that he describes his algorithm as using a weighted average in determining a pixel’s value across multiple stacked prints. However, if an image is blurry and low-detail from several generations of printing, it will be blurry in the average of these values as well. Worse, if you have a single high quality print stacked with several lower-quality prints, a weighted value would result in the lower-quality image. The only appreciable gain would be in removing the grain, which is what we see from his process.
I made a comment to this effect and even started an entire thread discussing a potential solution:
I would really like to do a test to see if this would actually work.