Topaz Labs recently released its all-in-one Photo AI suite that magically removes noise, sharpens, and increases resolution all at the single click of a button. But how does it compare with its own individual products, in this case DeNoise AI?
If you're not familiar with Topaz Labs' editing products, for a long while now they've been up towards the top of the tree for correcting noise in an image and sharpening as well. Until a short time ago, Topaz Labs' products were separate, in that you had DeNoise AI, Sharpen AI, and Gigapixel AI. The first two are self-explanatory in what they do, while Gigapixel AI adds resolution to an image, which can really help if you want to print an image at large sizes. You could (and still can) buy them all separately, or buy them as part of a bundle. If you bought them as a bundle, you still had to use them independently of each other. Very recently, however, Topaz Labs released Photo AI, which is an all-in-one software that removes noise, sharpens, and increases resolution with a single click. It sounds amazing, but how does it compare when used against its own individual products?
That brings us to this great video by Anthony Morganti, in which he puts the new Photo AI up against DeNoise AI, to see which one works better at noise removal. It's a fascinating experiment, simply because it's putting a company's products up against each other. Whatever the result, you can only deduce that there must be some form of cannibalization going on. If Photo AI comes out on top, does it make DeNoise AI entirely redundant? And if DeNoise AI wins, does it mean Topaz Labs has wasted its time on the new Photo AI? Give the video a look and let me know your thoughts. Personally, I like retaining the control that DeNoise AI gives me.
Please, f-stoppers: no software product "increases resolution" and you guys know this.
From the Topaz Labs website:
“Gigapixel AI helps you do a single thing very well: increase image resolution in a natural way.”
Photo AI:
“Maximize your image quality, on autopilot. Sharpen, remove noise, and increase the resolution of your photos with tomorrow's technology.”
A shame to see a slightly flippant response to a genuine point...I can increase the resolution of a 16MP image to 100MP image in Photoshop if I want but it doesn't mean it's going to have any extra detail.
It is absolutely worth mentioning the way this software adds resolution (i.e. uses AI to fill in gaps for you), otherwise people may get the wrong impression. But hey, I guess that's where the results of the comparison come in.
It comes down to this: does made-up detail count as additional resolution?
It's not exactly "made up" detail, it's more complex than that. With the technique they're using (essentially training this AI on thousands and thousands of low res to high res images) the software is going to get pretty dang good at approximating what detail would have been there in the first place. It's a fascinating technique and it does provide a real increase in actual detail (not just a bilinear upscale, for example). The noise plugins work in a similar fashion, and I've seen images shot at 32000 look like.. maybe 1600. It's actually insane.
Your point would have been valid a few years ago, but it's simply not the case anymore. You're working on old information.
I get that the added "detail" is better today. It's still fake, just more convincing. Basically, it's what's called "deepfake" in another context - very hard to detect.
So it's "detail" that the "photographer" neither captured nor created.
Whether you see this as a proper future for photography, or a loss of authenticity, is a personal thing. I think you can infer my attitude on this - I put it right up there AutoTune vocals.
But basically I just disagree with calling these added pixels "resolution". That term has a specific meaning in photography. It's a metric for what was captured - not for how the image was elaborated later.
They’re all good points you raise, Jim. But I think you’d be better off addressing Topaz Labs on their claims rather than Fstoppers. I just reported what Topaz Labs says on its websites regarding its software.
> But basically I just disagree with calling these added pixels "resolution". That term has a specific meaning in photography. It's a metric for what was captured - not for how the image was elaborated later.
This is 100% false. Resolution means the number of pixels in an image. It has absolutely nothing to do with the contents of those pixels or of that image. If you have a black square that is 1000x1000, and you make it 2000x2000 by copying every pixel one to the right and one down, you have added absolutely nothing to the image, no detail, no change, but you HAVE doubled the resolution.
In Lightroom it's a bit clearer as they're two separate functions under "Enhance" - you can "Enhance", which is AI/ML extrapolation of details that were not captured - WITHOUT touching the resolution of the image at all ... and separately you can increase the resolution which is literally just making the image *bigger* without altering anything more than that.
Ever since Photo AI came out that's all I use it is fantastic there's no use for the other three anymore. I run all my low light photos through it and they come out beautiful. This is a godsent program it saved me on a wedding I did when the camera stayed at 6400 ISO it removed all the grain or as we call it noise in the photos.
One setting I noticed as you were using Photo AI was Subject Detected - Auto. I have found that sometimes what Photo AI thinks is the subject I want it to work on isn't quite right. By Selecting Refine, you can see what it thinks the subject is and you can change it, or just say none. With none it operates on the entire image. I bring this up because it seemed that Photo AI did OK in some areas but left other area alone.
I have the full suite of all four apps. I am impressed with all of them, but I only use Photo AI when I need what it offers, It is indeed an amalgam of the earlier three, but it has some issues yet, one being that it is doing so much at once that it's slower.
I shamelessly shoot at 6400 and higher when needed, and I use half-frame APS-C cameras so things can get a little noisy. I just shot a gala at ISO 12,800. Huge room, hundred or more people, entertainers, speeches, awards...
And no frickin' light.
I have a flash and know how to use it, but I hate the look and would rather tame some noise than destroy the look of the ambient.
Processing, I culled down to the keepers and took one into Photo AI. It did a fine job but took a long time and I had 160 images to finish. Except for gnarly noise, they were otherwise fine, so I tried true and trusty DeNoise AI. That was all that was needed, and batched through the whole thing in half an hour or so.
I honestly hope that if Topaz ultimately intends to go all-in on Photo AI that they make it selectable. When a picture needs only good denoising and a little extra sharpening, DeNoise is still king.
The square crops emphasize the difference. #3 is the DeNoised version, while #4 is the final processed but NOT denoised, not in Lightroom or otherwise.
Camera, Sony A6400 with 18-105 f/4 zoom, 1/125 sec @ F/5.6, ISO 12,800.
CLICK BAIT - I click on an article that asks a question, implying that I can read the answer, BUT IT'S NOT HERE. That's a miserable way to treat your readers.
F-stoppers is largely a video aggregator, as in this case where the article teases a video, which might answer the question, or maybe just repeat the vendor hype. So, two levels of click-bait.
And yet we call this the "age of information".
The improvement is not enough to warrant me spending money on it. I can do as well with Camera Raw.