I planned to write a lengthy email giving Topaz Labs some feedback on their Topaz Photo AI app, but then I figured it would be too difficult to cover everything in an email.
So, here we are.
Fair Warning: this post might only interest Topaz and possibly a few people who own or are considering getting one of Topaz’s AI apps or their entire suite.
*** Unless interested in photo post-processing apps, best give this post a pass. ***
NOTE: This is in no way to be considered a criticism of any of these apps. They are my go-to apps for most of my post-processing, and I have nothing but praise for the company and the products. All I’m doing is pointing at something they might want to fix.
Right, here we go!
We begin with the above photo. It was taken with my then-Nikon D200 camera. Why that photo? No reason; it just happened to be the photo I opened to try out Topaz’s recent release of Topaz Photo AI.
Briefly, that’s an app that’s meant to evaluate your photo and automatically apply noise reduction and sharpening. I think it also does a few other things behind the scenes, but those are the current automatic functions. You then have the option to choose to enhance and/or enlarge the photo. Basically, the app incorporates Topaz Labs DenNoise AI, Sharpen AI, and Gigapixel AI.
For the above photo, Photo AI determined it needed Noise Reduction but not Sharpening, something I don’t necessarily agree with, but OK. (Note: as mentioned, I’m dealing with cropped versions of the photos so that I can show details.)
Note that the program applied the RAW noise reduction since this was a RAW image. Look at the license plate . . . not exactly an improvement from the original. The screenshot above shows the menu, but here’s the output file.
It’s not just the license plate. The headlight is also looking a little wonky.
Now, here’s what happens if I turn off the noise reduction and just turn on sharpening.
At this point, were I to switch on the noise reduction again, the photo would again degrade. The issue is with their RAW noise reduction engine . . . and it’s the same issue when using DeNoise AI.
Notice what happens when I switch to regular noise reduction in DeNoise AI . . . the issue goes away.
Here’s what happens when I use Sharpen AI . . . no issue because there is no RAW option.
Also, pay attention to how the dried grass looks when processed with RAW noise reduction versus ‘regular’ noise reduction.
Now, here’s the interesting thing . . . I get better results when using Gigapixel, even if I don’t enlarge the image.
I think — for that photo — Gigapixel produced the best results.
It makes me think the apps are not integrated with the same noise reduction and sharpening engines. Also, later, when you look at the slideshow, you will notice there are tonal differences between the files.
That’s an annoyance I mentioned before. When I choose to output the same as input (a RAW, or DNG file), I would expect all of the outputs to look the same and to retain the same information as the original RAW file, but that does not appear to be the case.
Again, it seems they should use the same output algorithms for each app.
Perhaps there’s some limitation I’m not aware of, but, again, a tad annoying, especially when I lose the in-camera color modes and whatever replaces them is not consistent across the board.
You might think it’s only one file, but here’s a file from my D100.
Look at the neck where the black transitions into the white. And the same below for DeNoise AI.
Here’s the DeNoise AI standard.
And Topaz Photo AI Sharpen without noise reduction
And Gigapixel AI
Again, the RAW engine noise reduction fares poorly in comparison, and I deem Gigapixel the better version of the above examples.
But, you might be thinking it’s because those are RAW files from old cameras. Remember this shot from last week, taken with my D7500?
If I just use Photo AI with sharpening and Enhance, this is what I get.
That’s not bad, and I have a small measure of control over the amount of sharpening and enhancement. I can probably use that with further tweaks, but some areas look a bit iffy.
Here are the rest:
Readers might note that you get slightly different results depending on which program you run first.
For most of what I shoot, I get better results running DeNoise AI first, then Sharpen AI. I then typically run the photo through Luminar AI, and then back to Lightroom for final tweaks.
I like what Topaz is trying to do with Photo AI. I would love a program that automatically ‘fixes’ my photos and makes them look great, or, at least, get me 75-85% there.
Currently, Photo AI doesn’t get me any closer to a good final product than using the original AI apps. Worse, in many cases, it degrades the original. Even when it does a decent job, I still think it needs more user-modifiable adjustments because the AI won’t know what I want, and if I can’t modify what it gives me, I don’t always get something I can use.
Again, depending on the photo, it can output pretty good results, but they need to fix the whole RAW noise reduction problem and ensure that the output is consistent across the apps (or explain why it’s not).
Finally . . . Gigapixel . . . here’s the version for the heron, and, again, I think it’s the best one of the bunch. I can take that and tweak it to end up with a pretty good version.
As a reminder, this was my output from the other day . . .
Here is the slideshow of the above, and the Gallery is at this LINK.
That’s it. This post has ended . . . except for the stuff below.
Note: if you are not reading this blog post at DisperserTracks.com, know that it’s copied without permission, and likely is being used by someone with nefarious intentions, like attracting you to a malware-infested website. Could be they also torture small mammals.
Note 2: it’s perfectly OK to share a link that points back here.