Topaz Photo AI verdict
$199 is a lot of money to pay for a simplified AI photo fixer and there’s not even a trial version, just an ‘unconditional’ money back guarantee. When it works, Photo AI is good, even spectacular, but the image and its problems have to fall within its window of fixability. Photo AI is also slow, over-aggressive with noise reduction and can only fix the right sort of blur.
+Sharpening can be spectacular. Can be
+Noise isn’t just reduced, it’s obliterated
+Blurry faces are magically enhanced. Kind of
-Very expensive for something that may help or may not
-Slow to install and to use
-Remove Noise doesn’t reveal much (or any) detail
-Sharpen can’t fix all blur types, despite the claims
AI is everywhere in photo editing software right now, from subject recognition and masking, to sky replacement, to noise reduction and detail recovery. We’re told AI can work magic where traditional photo editing science has failed. Either that, or it can achieve in seconds what it might have taken us hours to do manually. It can put expert adjustments in the hands of amateurs.
It’s not always easy to figure out how much of this is snake oil and how much is genuinely effective and Topaz Labs – and this is just my opinion – steers carefully down the middle. Its marketing can sound like snake oil, its results can sometimes be impressive.
Topaz Photo AI takes three of its separate programs – DeNoise AI, Gigapixel AI and Sharpen AI, and bundles a simplified version of each into Photo AI, which will analyse your photos and fix what needs fixing with the magic of artificial intelligence, deep learning and neural networking.
Topaz Photo AI can automatically apply a series of adjustments using its AI technologies to remove noise, sharpen details and fix blur, reveal detail in faces and – if you need it – upscale images to much larger sizes.
Let’s start with the noise reduction. It is spectacularly effective, in that it will leave no vestige of noise whatsoever. But it doesn’t really leave a lot of detail either. It doesn’t create that smudgy watercolor effect you used to see in cheap point and shoot cameras, but it does leave you with disappointingly ‘glassy’ detail.
I thought I’d check it against DxO’s DeepPRIME and DeepPRIME XD processing, and it was like night and day. The Topaz results are not in the same league for detail recovery and yes, I did try all the settings. And can I point out that a Topaz Photo AI license costs almost as much as PhotoLab 6 Elite?
I don’t often need to extract the best quality from ultra-high ISO images, but if I did, I wouldn’t use Topaz Photo AI.
The Sharpen feature is a different story, in a way. I found that on images that were slightly soft, perhaps from very mild camera shake or diffraction from small lens apertures, it was spectacularly effective. I won’t deny it. Yet on images with obvious camera shake or poor focus it didn’t do much at all except draw attention to the issue.
The face enhancement can be quite remarkable or can be quite ‘plastic’. I can see it being useful for people shots where somebody important is slightly out of focus either in the foreground or the background. Topaz Photo AI brings out facial features in a fairly obvious ‘processed’ way, but it will definitely please family, friends and fussy clients.
The upscaling feature is where, I admit, I have a problem. As I found in my Gigapixel AI review, the results are very impressive, though more with textures than straight lines and man-made objects. My problem here is why you would need it.
Good reasons include lost originals where you only have a low-res copy, images sent by family, friends or clients who don’t know how resolution works, and component images you need to upscale for multi-layer composites.
But I can’t help thinking of the bad reasons, like low res images scraped from the Internet that people want to use as their own.
Either way, I just feel that upscaling software is a sometimes effective solution to a problem you shouldn’t have.
There are some things that Topaz Photo AI does that I think are genuinely remarkable. But there’s plenty more that isn’t. I particularly dislike its marketing. It seems to suggest you can be a pretty hopeless or unlucky photographer and that buying this software will fix it. To be honest, I think if you do buy it for $199, you’re the one who’s been ‘fixed’.