Since I was dealing with people I've known most of my life, my bar for accuracy in facial reconstruction was very high. For those not familiar with the subject the work may look acceptable, but it may be hideous to those familiar with the person since our brains are very attuned to subtle facial cues and immediately see something is off. In particular, the facial reconstruction is pretty amazing.Īs you may know, reconstructing faces can be challenging. I have to say the product is much improved. A free update to Gigapixel was available to me since I had purchased it a while back, so I ran the update. Since the low resolution image was a result of algorithmic deterioration, the ML algorithms did a better job reversing the process on those than real-world scans and source materials.įor this project, I had some desirable photos that were just too blurry.

My opinion was that, for their demos, Topaz may have taken high res originals and made them low resolution through downsizing and JPG compression, then used those as their source material to have the machine learning software make a high resolution result. I bought it because their demo was amazing but, in practice, I found it did not work as well on actual low resolution photos I had, but read on. A family member recently passed and I was working with a lot of old blurry family photos, sometimes damaged, to assemble a memorial photo collage of their life.Ī few years ago, I bought Topaz Gigapixel AI. I'm a photo restorer and photography enthusiast who uses Photoshop and related 3rd party plugins.

Gigapixel ai reddit