In my recent review of Topaz Labs’ Photo AI, I came away more than a little impressed by its AI-powered Enhance Resolution function. Although I’ve used competing resolution-boosting tools like Adobe’s Super Resolution and ON1 Resize AI in the past, I was nevertheless surprised by the usability of Photo AI’s resolution enhancement. Even when processing photos with quite low resolutions it delivered usable enlargements, with the only Achilles’ heel being the results when shots included recognizable logos or barely-readable text.
My time with Photo AI made me extremely curious to see how the program from which Photo AI drew its resolution-enhancing algorithms, Topaz Gigapixel AI, might perform given its rather greater selection of controls. And hot on the heels of an update promising even better results, I was keen to learn how it compared both to its nearest rivals. Let’s take a look at how it stacks up to what Adobe and ON1 have to offer.
 |
| The Super Resolution tool in both Adobe Camera Raw and Lightroom is an on/off switch, with no user controls whatsoever available. |
Three different approaches to UI and complexity
It’s worth noting up front the most immediately obvious difference between the trio of AI options: their interfaces and overall capabilities.
With Adobe, Super Resolution is literally just an on/off switch that lets you double the linear resolution – that is to say, the width and height – of an image (or, put another way, quadruple their megapixel resolution). There’s no user control whatsoever beyond that, and your results are rendered out to a DNG file regardless of the source file format.
Although it’s technically possible to process either Raw files or JPEGs using Super Resolution, I’ve compared numerous enlargements side-by-side and see no real-world reason to use the tool on JPEGs. With JPEGs I could see no meaningful improvement over resampling with Adobe Photoshop’s Preserve Details 2.0 algorithm. Super Resolution’s results did tend to be just fractionally crisper, but at the cost of being more prone to pixelation.
To my mind, Super Resolution can be considered an effective tool only for Raw files, where the difference in image quality over other upsampling methods is clearly visible.
 |
| Topaz Gigapixel AI has by far the most complex interface of the trio and provides the most control, but it’s still simple and easy to operate. |
Topaz Gigapixel AI, meanwhile, occupies the opposite end of the user options spectrum. Although it can certainly do its task without user intervention, the expectation is that you’re going to tweak and tune the settings for best results based on subject matter. You can choose from one of six different AI models specific to different types of imagery, and you can then adjust sliders to suppress noise, remove blur, and (for some AI models) fix compression artifacts.
You can scale by width/height or by choosing an enlargement ratio of up to 6x, although if you go any further there’s a bit of a gotcha. Gigapixel AI will only use its machine-learning algorithms for that initial 6x enlargement before falling back to more traditional resampling methods for the remainder.
It also gives you almost the same face recovery tool as in Photo AI; at least its strengths and weaknesses are much the same as in that product, so I’ll encourage you to read my earlier review. I should note quickly, though, that it’s an all-or-nothing proposition in Gigapixel AI, as you can neither see which faces have been detected nor disable the effect on a per-face basis, only for the image as a whole.
 |
| ON1 Resize AI – seen here hosted as part of the Photo Raw 2023 app instead – strikes a middle ground, offering a lesser degree of user control than Gigapixel AI. |
Finally, ON1 Resize AI assumes something of the middle ground. I should note here that I’m using the version of the tool built into ON1 Photo Raw 2023, which I also recently reviewed, rather than the standalone application. ON1 describes it as being ‘fully integrated‘ into Photo Raw, however, so its functionality and underlying algorithms should be identical.
With ON1’s offering, you can choose a scaling factor of 6x or beyond. (Just as with Topaz Gigapixel AI, the actual upper limit will depend upon your source image.) You can also select one of two interpolation methods (Standard or Faithful) and have some control over sharpening and grain generation, but there’s no way to tweak results beyond that.
Ground rules for the comparisons
To draw my conclusions I tested all three applications with dozens and dozens of Raw and JPEG photos from a wide variety of digital cameras, ranging from ancient to quite modern. For this article, though, I’m going to illustrate just five sets of comparison crops.
We’ll start by comparing JPEG and Raw enlargements from low-res and high-res originals. Finally, for the fifth comparison, we’ll see how an image survives first being downsampled to a lower resolution and then upsampled back to its original resolution with each technique, to give a sense for how much is lost along the way.
A version upsampled in Photoshop with the Preserve Details 2.0 filter will serve as the baseline.
 |
|||
| Adobe Preserve Details 2.0 | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (HQ mode) |
| Full sample image processed to 2x original resolution with each technique. Mouse over the titles to switch images; click the images to see originals. | |||
Upsampling a low-res JPEG original
We’ll start off with a JPEG shot taken with my 3MP Canon PowerShot G1 way back in November 2000 on a trip to New York City. Below, let’s take a look at a few 100% crops:
 |
| The low-res JPEG image: 100% crops below made from this image. |
 |
|||
| Adobe Preserve Details 2.0 | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (Low Res mode) |
As you can see, there’s nearly no difference between Preserve Details 2.0’s upsampling algorithm and Super Resolution for this JPEG crop, save for a very, very slight improvement in sharpness.
ON1’s Resize AI algorithms help a bit more, making the result rather more crisp, but the ship’s name looks a bit pixelated and the cable above and to its right looks a bit unnatural.
Topaz Gigapixel AI does the best job overall, rendering the text quite cleanly. It does lose a very faint diagonal banding which is actually subject detail in the shadow beneath the name, though. Also, a few of the automatically-tweaked faces look slightly unnatural.
Here’s another crop:
|
|||
| Adobe Preserve Details 2.0 | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (Low Res mode) |
Again, there’s next to nothing in it when comparing Adobe’s Preserve Details 2.0 and Adobe’s Super Resolution for this JPEG image.
ON1 Resize AI does a much better job than Adobe again, but parts of the buildings have a slightly stippled, pointillistic look. It does hold onto some surprisingly fine detail in the low-contrast windows at center left very nicely, though.
For this crop, I’d have to say Topaz Gigapixel AI looks the best overall and it’s certainly the crispest. Seen really close up close like this, though, it does feel slightly oversmoothed by default. Also some of the lower-contrast windows have nearly vanished altogether.
|
|||
| Adobe Preserve Details 2.0 | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (Low Res mode) |
Once more, Adobe Super Resolution shows negligible advantage for JPEG files.
ON1 Resize AI is much crisper, but the water feels a bit unnatural and overprocessed when viewed very closely.
Gigapixel AI does by far the best job here, mostly yielding quite smooth and natural-looking ripples in the water. What little artifacting remains can be seen in the most turbulent water immediately behind the bow wave.
It’s worth remembering that both Resize AI and Gigapixel AI can go beyond the mere 2x enlargement of Super Resolution, too. Here, let’s compare a couple of 4x and 6x crops with both programs:
 |
||
| Adobe Preserve Details 2.0, 400% | ON1 Resize AI (Standard mode, 4x) | Topaz Gigapixel AI (Low Res mode, 4x) |
At 4x, the AI’s guesswork starts to mean that you won’t want to look too closely, as it’s easy to notice unnatural artifacts from either ON1 or Topaz if you’re looking for them. There’s no denying that both images are more useful than the heavily-blurred version that was resized by traditional methods, though.
And if you’re making very large prints that won’t be viewed right up close, the AI-invented subject “detail” certainly helps give the illusion of crispness to the eyes from a bit further back.
 |
||
| Adobe Preserve Details 2.0, 600% | ON1 Resize AI (Standard mode, 6x) | Topaz Gigapixel AI (Low Res mode, 6x) |
At 6x, the traditionally-resampled image is blurred far beyond usability. Both AI-authored versions, while containing some ugly halos and artifacts, could again still be more usable so long as your viewers don’t get too close. You can see here, incidentally, that the faint banding Topaz removes is actually a pattern in the joists beneath the roof.
Upsampling a low-res Raw original
Next up, a .CRW Raw shot captured in 2002 with the same 3MP Canon G1:
 |
| The low-res Raw image: 100% crops below made from this image. |
 |
|||
| Adobe Preserve Details 2.0 | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (Low Res mode) |
Interestingly, this shot shows a bigger difference between the Adobe crops than with the JPEG, but things have progressed in the wrong direction. The Super Resolution shot is actually slightly softer than that scaled using the Preserve Details 2.0 filter!
ON1’s defaults produced a very noisy image here, something I could have cleaned up manually for a better result. The noise has stymied the AI, though, with unsightly artifacts everywhere, and neither Adobe nor Topaz needed manual intervention to deal with the noise.
Topaz Gigapixel AI does by far the best of the bunch. Its image feels a touch oversmoothed, but it shows a good bit more detail in the hair and logo text alike.
|
|||
| Adobe Preserve Details 2.0 | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (Low Res mode) |
Again, Topaz has a clear edge while ON1 fights the noise and Adobe fights its self-created blur.
|
|||
| Adobe Preserve Details 2.0 | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (Low Res mode) |
And once again Gigapixel AI looks best. But look at that T-shirt to see what we mean about the image feeling overly smoothed-out!
Upsampling a high-res JPEG original
Next up, a much higher-res 20.1-megapixel JPEG shot in 2020 with the Panasonic Lumix FZ1000 II:
 |
| The high-res JPEG image: 100% crops below made from this image. |
 |
|||
| Adobe Preserve Details 2.0 | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (HQ mode) |
There’s still not that much difference between Preserve Details 2.0 and Super Resolution in a JPEG image, even at a much higher source resolution. With that said, both Adobe algorithms come closest to their best in this shot. ON1’s algorithms leave some speckled artifacts along the edges of the flaking paint, while Topaz Labs turns in the best result but looks a bit overly smooth.
|
|||
| Adobe Preserve Details 2.0 | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (HQ mode) |
Gigapixel AI’s algorithms definitely do the best job with the grille, completely erasing the noise artifacts which ON1 Resize AI unintentionally emphasizes.
|
|||
| Adobe Preserve Details 2.0 | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (HQ mode) |
As you can see in the text on the tire, ON1 Resize AI does give the appearance of a bit more sharpness, so if you don’t look closely enough to be bothered by the artifacts it’s a bit crisper. Topaz Gigapixel AI gives the most pleasing result again.
Upsampling a high-res RAW original
Now let’s continue to an even higher-res 51.4-MP .DNG Raw shot in 2014 with the Pentax 645Z:
 |
| The high-res Raw image: 100% crops below made from this image. |
 |
|||
| Adobe Preserve Details 2.0 | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (HQ mode) |
There’s finally a more noticeable difference between Adobe’s Preserve Details 2.0 and Super Resolution when working with a high-res Raw file. The text picks up some slightly unsightly haloes, but it also appears a little crisper. Both ON1 and Topaz show some of the usual issues AI suffers when dealing with text. On balance, Topaz’ result is a little more pleasing here, though.
|
|||
| Adobe Preserve Details 2.0 | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (HQ mode) |
Resize AI gives a somewhat unnatural look to the girl’s face. It also makes the shirt of the man in the foreground look almost cartoonish, as does Gigapixel AI. They both manage to crispen up the slightly bokeh-blurred tree and people in the background more than Adobe Super Resolution. There’s a fairly noticeable improvement for Super Resolution over Preserve Details 2.0 in the tree trunk, though.
|
|||
| Adobe Preserve Details 2.0 | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (HQ mode) |
It does depend very much where you look in the image as to what you’ll note. Topaz Gigapixel again looks best overall and does quite nicely in this crop, but in other parts of the foliage Resize AI looks better.
The downsampling test
Finally, we’ll try something a bit different for the last test. Here, we’re taking an image shot with the Pentax 645Z again, but first I’ve downsampled it to half its original width and size using the Bicubic Sharper algorithm. I’ve then saved that out as a JPEG and then used each program to reupscale it again. This can then be compared to the original image to see how much quality has been lost in the process.
 |
| Prior to downsampling and re-upsampling: Shot in 2014 with the 51.4-megapixel Pentax 645Z. |
 |
|||
| Original image prior to downsampling | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (HQ mode) |
It’s really obvious comparing against the original image that Adobe Super Resolution is boosting the noise levels much more than ON1 Resize AI. It’s also notable just how smoothed the image from Topaz Gigapixel AI is. With that said, Topaz also manages to make the text look most like in the original, where both rivals make it look too bold.
|
|||
| Original image prior to downsampling | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (HQ mode) |
Again, the higher noise levels of Adobe Super Resolution are obvious here. Interestingly, for this crop Resize AI seems to look the most smoothed, while Gigapixel AI looks the crispest, especially in the pattern and text on the headlamp lens.
|
|||
| Original image prior to downsampling | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (HQ mode) |
It’s interesting that Gigapixel AI sometimes manages to look crisper than the original image despite having been made from just 1/4 of its data. Again, the overly smooth look makes this seem almost painting-like, and the bokeh-blurred white grill at the left of the image looks especially funky after the smoothing. But that silver metal bar in the foreground is really clean and crisp! ON1 does a pretty great job also, while Adobe Super Resolution shows the smallest improvement but lots more noise.
Performance
Perhaps my most surprising discovery in writing this comparison was by just how much the three different programs varied in their speed. All timing was performed on a mid-range, 2018-vintage Dell XPS 15 9570 laptop with 2.2GHz hexa-core processor and running Windows 10 21H2.
For my comparison, I batch-processed the same five images from which I’ve drawn all of my 100% crops throughout this article. With Adobe’s Super Resolution, it took a little over three minutes to render three Super Resolution DNG files and another minute or so to render them to JPEGs, for a total of just four minutes.
Rendering the exact same five images at the same resolution took a whopping 48 minutes with Topaz Gigapixel AI. That’s twelve times longer, but the crazy thing is that it was still way faster than the last-place entrant. ON1 Resize AI took a stunning 97 minutes, more than twice as long as Gigapixel AI and 24x longer than Adobe Super Resolution.
 |
|||
| Adobe Preserve Details 2.0 | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (HQ mode) |
| Full sample image processed to 2x original resolution with each technique. Mouse over the titles to switch images; click the images to see originals. | |||
Clearly, the lesson to be learned here is that there’s a reason Adobe Super Resolution trailed in every test. It’s not setting out to throw vast amounts of processor power at the problem like its rivals do. You thus have to consider that its more modest improvements are available to you an order of magnitude more quickly and easily than its rivals.
The other takeaway is that if you want the best possible results, you’ll need to be willing to wait to get them – especially when working with large image batches or high-resolution source images.
Conclusion
For my money, the clear winner here overall is Topaz Gigapixel AI. While it’s a bit prone to over-smoothing it generally gives clearly better results than its rivals. And that smoothing can easily be dialed back manually if, like me, you prefer a less-plasticky look.
The big disappointment of the trio was, surprisingly, Adobe. While there’s a more noticeable difference to be found in upsampling from higher-res source photos, in the lower-res shots where it’s potentially most useful, Super Resolution proved to be basically indistinguishable from the Preserve Details 2.0 filter.
 |
|||
| Adobe Preserve Details 2.0 | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (HQ mode) |
| Full sample image processed to 2x original resolution with each technique. Mouse over the titles to switch images; click the images to see originals. | |||
By way of comparison, Preserve Details 2.0 debuted in 2017. Yet despite arriving almost four years later, Adobe Super Resolution offers surprisingly little added benefit and trails the competition by quite some way. But once again, Adobe’s algorithms are much faster than its AI rivals.
Nor is Topaz Gigapixel AI completely perfect. While I definitely preferred its results over those of its rivals in terms of crispness and detail, it does have an all-too-frequent tendency to miss by some way on color or exposure, and there’s no way to correct this in-app.
For my money, the clear winner here overall is Topaz Gigapixel AI.
You can, however, render to a DNG or just use Gigapixel AI as a plugin instead, and then fix any color/exposure issues in the other app. Hence, I don’t see this as a huge issue, even if I think Topaz Labs definitely needs to look into it.
I’d also like to see the company add both the ability to disable face recovery on a per-face basis and some basic white balance/exposure controls in standalone mode at the least, even if not also when functioning as a plugin.
As for Resize AI, while ON1 couldn’t quite keep up with Topaz, it too turned in a significantly better showing in terms of crispness than Adobe Super Resolution. It was the slowest of the bunch by far though, and lacks Gigapixel AI’s flexibility while equaling it on price.
 |
|||
| Adobe Preserve Details 2.0 | Adobe Super Resolution | ON1 Resize AI (Standard mode) | Topaz Gigapixel AI (HQ mode) |
| Full sample image processed to 2x original resolution with each technique. Mouse over the titles to switch images; click the images to see originals. | |||
The good news is that if you own ON1 Photo Raw 2023, you already have Resize AI integrated into that app! Likewise, Super Resolution is free for anyone who’s already using Adobe Photoshop or Lightroom. The not-so-good news is that both Gigapixel AI and Resize AI are a little on the pricey side – you’ll be spending around $100 for a perpetual license for either of those AI apps.
To put that in perspective, it’s only $40–50 less than much more fully-featured apps like Skylum Luminar Neo or DxO PhotoLab. It’s also enough to cover a ten-month, one-terabyte Lightroom subscription from Adobe.
Not everyone can justify that much for a single-purpose plugin, even if it’s one that can also be run standalone. If you regularly find yourself needing to upsample your images, though – and especially if you shoot subjects like landscapes, wildlife and nature where natural detail predominates – then your money would be well-spent on either Topaz Labs or ON1’s products, both of which handily outperform Adobe Super Resolution.
Of that pair, if you don’t have a workflow need which would make the decision for you, I’d recommend opting for Topaz Gigapixel AI. It will give you the best results most of the time, and do so while providing the maximum flexibility.
All processing was done using the latest publicly-available Windows versions of the various apps as of this writing. (That’s Adobe Camera Raw v15.1.1, ON1 Photo RAW 2023 v17.0.2.13102 and Topaz Gigapixel AI v6.3.3.)
Author:
This article comes from DP Review and can be read on the original site.
