In late April I was finally able to declare my photo digitisation project “completed”.
It took me exactly one year to clean up, research, name, and sort over 7,000 scans and photos which ultimately made it into my personal “digital photo album” (if that’s what a stringent directory structure can be called) – and that’s excluding the time spent scanning them in the first place, or culling the ones that didn’t make it.
At this point it would be remiss not to mention AI photo enhancement software.
Despite the recent incredible developments in artificial intelligence and image generation, I remain steadfast that AI still has no role in the workflow for digitising personal snapshots on prints, slides or negatives. While I obviously made basic edits like cropping, or adjusting brightness, contrast, white balance and colours so that the viewer can actually see what’s going on in a photo, my experiences with AI services (read: face enhancing) have done nothing but confirm a phenomenon that’s already been termed “identity shift“.
Make no mistake when I say that my mind is utterly blown by the “imagination” and sheer possibilities that GANs offer — if you can look past the fingers and the hallucinations.
Although they have ushered in a renaissance with bright new avenues (as well as dark alleyways) in creativity, data presentation and, indeed, even “inspiration”, all my personal observations have confirmed thus far is that AI is great at one thing: remixing.
It fails at restoration. Filters and fun effects have existed for ages – but distortion is not restoration. Alteration is not enhancement.
Smoothing away wrinkles and blemishes might be the ambrosial cash cow of the cosmetics industry but it has no bearing on capturing realism and an unglamorous truth. Everyone wants a proper “hero shot” but sorry, creativity is not reality. It’s pure moonshine to claim that missing pixels can be sucked out of thin air!
As could be expected, many questionable online operations that promise to fix your photos have popped up in recent times. You’ll have to forgive my scepticism about those which resort to fake testimonials by people who do not exist.
Artificial intelligence is no match for natural stupidity.
And as an aside, why, oh why are these dubious product reviews always “written” by people with English/American-sounding names? It’s clear there’s typically a bunch of Indians or East Asians behind the operation. Is the trope over there that white folks have lots of cash to spare, exude an air of authority – or are gullible? Mind you, couldn’t this be described as a form of racial profiling? Best not to go there…
Bias within AI is a notoriously contentious issue, and it fans out in several directions: one utility I tested recently did a remarkable job of turning a North African woman into a Nordic blonde while another restoration tool by Tencent assumed I must’ve been a bubbly Chinese boy with a disfigured chin. It’s a load of fun and entertainment but there is a difference between fact and fantasy — especially when a person’s eye colour is changed! You recognise the photo but not the person in it.
With that in mind, there are utilities like Topaz Photo AI which deliver believable results. On closer inspection, though, what they ultimately do is mutate human faces into creepy residents of uncanny valley while blissfully ignoring everything else in the picture.
Heaven forbid people should have earrings, or hands or hair obscuring part of their face!
Indeed, attempting to upscale and restore anything other than a face can lead to very strange results as well as a casual conclusion that human faces are easy to define, recreate and therefore surveil — or they just happen to form the largest portion of the training data.
Either prospect is equally alarming.
All photos and screengrabs by Herby Hönigsperger unless specified otherwise.