[robgski]

It's just a very advanced algorithm, not intelligence.

I'm no expert, but the algorithms are a method by which AI is achieved and packaged for commercial use on a consumer grade PC, IIRC. Sure, there are other AI techniques like neural nets and unsupervised learning, but they require more infrastructure, and expertise to bring about.
Also, any machine learning or AI is only as good as the data set it's trained on, I assume the data set emphasis was on both facial features as well fine detail like hair or feathers.

The challenges of computer vision distinguishing just what the blurry form is, distinguishing it from both a blurry background, as well as shadows. Impressive, IMO. Not only that but being able to provide feedback to Topaz while doing processing is a good feature.

FWIW, In the Topaz AI user group I posted a few examples of photos I've processed via various TopazAI products. I need to post more but I've been swamped with other tasks.
I was very impressed with GigaPixel being able to take a scanned 3 x 5 photo and scaling it up to at least the size of my 27 " computer monitor with impressive detail

[dms]

I use AI sharpen a lot for motion blur mostly (and sometimes for missed focus), in my theatre photography, but nevertheless I have serious misgivings about it (but it is just so good usually). But you need to check what it does, in one case it added a 6th toe (or a finger--I have forgotten now--but I wrote about it in a post here).

I agree to call the software AI is incorrect--while Topaz may have used AI to develop the algorithims in the software, the software does not learn/improve from our using it--so it is not AI.

[Fenwoodian]

.
Topaz Sharpen AI, Topaz DeNoise AI, and Topaz Gigapixel AI all sharpen and denoise images. There's only one major difference between the sharpening/denoising features in these three different softwares.

That difference is that Topaz Sharpen AI is the only one that has masking and local sharpening and denoising. I only sharpen/denoise locally. While I own all 3 of these Topaz products, when it comes to using them for sharpening and denoising the only one I ever use is Topaz Sharpen AI.

Also, Sharpen and Denoise have always had (and continue to have) FREE upgrades. With the recent Gigapixel AI upgrade, Topaz has just now started to charge for upgrades to Gigapixel AI.

So, if were going to buy them for the first time, I'd only purchase Topaz Sharpen AI. Why? First, it's the only one to offer local adjustments and masking. Second, upgrades are free.

[Mooncatt]

That difference is that Topaz Sharpen AI is the only one that has masking and local sharpening and denoising.

DeNoise has masking and local adjustments as well. Gigapixel obviously wouldn't, because how does one enlarge only selective parts of a photo?

I only sharpen/denoise locally. While I own all 3 of these Topaz products, when it comes to using them for sharpening and denoising the only one I ever use is Topaz Sharpen AI.

Sharpen and DeNoise have the functions of each other, but I've found them to specialise in their named function. Sharpen can apply noise reduction, but not as well as DeNoise. DeNoise can apply sharpening, but not as good as Sharpen.

Also, Sharpen and Denoise have always had (and continue to have) FREE upgrades. With the recent Gigapixel AI upgrade, Topaz has just now started to charge for upgrades to Gigapixel AI.

Their licencing program is a purchase gives you one year of upgrades, and perpetual use of whatever the last update version is during that year without further payment. DeNoise and Sharpen do not have free perpetual updates.

[biz-engineer]

they'll analyse the scene and actively replace chunks of it based on what they "think" the content should be.

Then you should be very upset about your own brain, because re-building own reality is exactly what human brains are doing all the time.

if you don't push them enough, often the difference between the AI enhancements and traditional raster image processing enhancements is less impressive.

That's because you aren't aware of the difference between perception (acutance) and accurate reproduction of detail. The goal of those AI algorithm is NOT the accurate reproduction of detail, but it is to increase the perception of image quality in normal viewing conditions (print or display crop). The eye is not able to tell if pixels are fake or genuine above 200 PPI, it's only able to feel if there is some detail or not, but not able to distinguish if the tiny details are genuine or fake. Let's consider you print two photographs of the same subject, equivalent exposure (equivalent depth of field etc..), one is photographed with a 645Z (50Mp) and the other one photographed with a K3 III (25Mp). The 645Z image is sharpened with unsharp-mask various radius, printed at 300 PPI, the K3 III image is upscaled to 300 PPI (+50%) and sharpened with Topaz AI. You won't be able to tell the difference between the two prints. But the difference in shooting with one camera or the other is huge: the K3 III gives you 12 FPS and 101 AF points, can capture images you will never be able to capture with a 645Z.

[BarryE]

Like Mike (OP), I too will use these tools, occasionally and with reservations.

The artifacts in the last example with the woman's glasses/eyebrows and sandals (earlier example) demonstrate to me the issue - the AI knows what a face looks like but random (to it) objects it can only guess.

[mlag]

Like Mike (OP), I too will use these tools, occasionally and with reservations.

Also using them with care as with any other tool or presets...

AI or not, it remain tools as any other, serve a purpose and need to be applied carefully when needed. Presets are nice but you might want apply improvements.... even something basic as a AWB on the camera needs sometimes an override.

Same for these topaz tools, sometimes nice , sometimes tuning required, sometimes discarded as useless for what I judged usefull.


I would start to feel uncomfortable if AI is outsmartening me under all circumstances...that would mean it can read my mind or that I am completely predictable and boring or that all humans think/like/behave identical...

But for the time being, AI means here they have processed/analysed a large number of pictures, of a large number of typical subjects, and created an automated sharpening/denoising tool to yield results/pictures pleasing a large number of people...which works in many cases but not always.... The final judgement is still human...as it might never cover all subjects, in all circumstances to everyone's taste.

[Rondec]

My biggest problem with these sorts of software packages is that images processed through them end up having a sameness to them. It isn't the end of the world, but when a photographer falls in love package and runs most of their images through them -- particularly when they get a little aggressive with the sliders, you can tell -- and not in a good way. I think Tony says that the best use of them is where you have had to do a heavy crop of something and you want more pixels than you have available. Of course, there's no free lunch, but at the same time, they look pretty decent for what they are.

[robgski]

My biggest problem with these sorts of software packages is that images processed through them end up having a sameness to them. It isn't the end of the world, but when a photographer falls in love package and runs most of their images through them -- particularly when they get a little aggressive with the sliders, you can tell -- and not in a good way.

Isn't that true for any processing software?
Like any other labor saving device, this is designed to take what I call the "necessary drudgery" burden off the human in the processing workflow, so the human can spend more time on the creative side of post processing. I use TopazAI Sharpen, DeNoise, and even GigaPixel as middle steps, not the end of the process. For someone who has not devoted hours and hours into learning PS or LR techniques, I find it to be a boon.

[Michail_P]

There are limitations, so what's new? There are always boundaries. Respect the benefits, recognise the flaws , learn and adapt. AI still has some way to go. Things will improve, maybe rapidly. For now it's only emerging.

[biz-engineer]

There are limitations, so what's new? There are always boundaries.

Enormous distortions in judgement of imaging these days. People zoom in at whatever magnification is possible, 100%, 200%, 300% , completely losing sight of how the image would look like as a paper / canvas print at its final size. The wrong judgement: never sharp enough at 200% on electronic displays, and yet never make large prints (or never make prints at all..).

Time to plan and travel to new location, use the camera, get off the internet hook.

[Rondec]

Isn't that true for any processing software?
Like any other labor saving device, this is designed to take what I call the "necessary drudgery" burden off the human in the processing workflow, so the human can spend more time on the creative side of post processing. I use TopazAI Sharpen, DeNoise, and even GigaPixel as middle steps, not the end of the process. For someone who has not devoted hours and hours into learning PS or LR techniques, I find it to be a boon.

How big are you printing? I find using a modern camera (even the K-01 qualifies), if I do a decent job framing and use a good lens then I don't need any special software to print at the sizes I use. I'm still using Lightroom 6.x and it is adequate for my purposes.

The advertisers would tell you that you can replace a larger format camera with these products, but I don't think that is true either. But certainly there is nothing wrong with using them.

[texandrews]

I have used Gigapixel AI to great effect in a museum setting. Like any tool, it requires a thinking user.

[Dartmoor Dave]

Let's consider you print two photographs of the same subject, equivalent exposure (equivalent depth of field etc..), one is photographed with a 645Z (50Mp) and the other one photographed with a K3 III (25Mp). The 645Z image is sharpened with unsharp-mask various radius, printed at 300 PPI, the K3 III image is upscaled to 300 PPI (+50%) and sharpened with Topaz AI. You won't be able to tell the difference between the two prints. But the difference in shooting with one camera or the other is huge: the K3 III gives you 12 FPS and 101 AF points, can capture images you will never be able to capture with a 645Z.

What method are you using to upscale the K3 photo in this example? Up to (and including) Lanczos resampling your example might be true. But if you are using Topaz Gigapixel for the upscaling then artefacts could be generated that might result in visibly false information. I've seen enough of those artefacts myself from Gigapixel to know that they can ruin the upscaling results.

It's not for me to speak for @BigMackCam, but my reading of his original post is that it's things like the false textures that Gigapixel can produce that he's concerned about.

[Dartmoor Dave]

Then you should be very upset about your own brain, because re-building own reality is exactly what human brains are doing all the time.

Somewhere, Bishop Berkeley is applauding you. Personally, I'm a Peircean pragmatist.