If you have low-resolution, blurry, and/or noisy photos, it is very possible that Topaz Photo AI is a tool for you. We are talking about both amateur and professional photoshoots. The subject matter is also very broad – it works with photos of people, nature, macro, etc.

Update: April 2023. I decided to check if anything has changed since the premiere and swapped the photos that are now much better with those edited by the current version of the program. First of all, very strong upscaling of photos with small details close together works much better (this is shown perfectly by the example of the tree without leaves). Face recovery better blends sharp areas with the rest of the photo, and the masking has been fixed.

I will start the presentation of Photo AI examples with people, and then mix them with all other types of photos. I’m uploading a photo to the program, and it tries to improve its quality:

Topaz Photo AI – denoise, face enhancement, and sharpening functions enabled, upscaling disabled.

The noise has been removed, sharpness has been improved, and the overall image has much more detail. It’s best seen by zooming in on the face that is furthest out of focus:

Topaz Photo AI – zoom on the face from the previous frame, upscaling disabled.

The original has a much higher resolution than the one on the blog (even in this small cropped area of the face), but here I am inserting reduced images so that upscaling doesn’t destroy the sharpness. The effect will still be difficult to see on phone screens, but on a monitor, you can appreciate what artificial intelligence can achieve today (and what it cannot do, because that will also be discussed). This will probably be one of the photos that I will follow the development of algorithms in subsequent updates.

Author: Agnieszka Majewska

Photo AI is three programs in one.

For years, Topaz has had great AI-based tools for enlarging photos (Gigapixel AI is even the best), denoising (DeNoise AI), and sharpening photos (Sharpen AI).

Topaz realized this and created the Topaz Photo AI tool, which enlarges, sharpens, denoises, and enhances faces, all in one program, which is also a Photoshop plugin.

Author: Przemysław Szklarski

The face enhancement function does not work on a simple on/off basis (although there is also a toggle for that). It also allows you to select the characters from which you want to enhance the faces, so I simply indicated only one person in the foreground to keep the depth of field natural. This photo mainly illustrates how denoising works, so it’s worth taking a closer look:

Author: Przemysław Szklarski

Here, in some places, denoising has been skipped (update: current version makes it better). I must admit that I have very little contact with noise as I have been taking photos with native ISO or close to it, and the photos sent to me for retouching are usually the same. However, I have heard mixed reviews about Topaz’s denoising tool (DeNoise). Some people have pointed out NoNoise as a better alternative, while others believe that Topaz’s DeNoise is the better tool. It’s probably largely a matter of the photos and the moment in which they were compared, as progress in AI is rapid. I haven’t used DeNoise yet, only denoising from Photo AI, which is based on it, but has fewer settings and I’m demonstrating its effects. I prefer to have Photo AI for almost everything, and dedicated tools for specific tasks.

Author: Przemysław Szklarski

Author: Paweł

Luminar Neo is not an alternative.

It may seem that Luminar Neo has similar functions. I will most likely write a separate article about it since I have it along with all the add-ons, but for correcting photos with missed focus, etc. Luminar is not suitable, or at least not in its current version. It’s a bit like comparing Lightroom and Photoshop – both are used for photo editing. Luminar Neo has a lot of features that Topaz Photo AI doesn’t have – they are simply two completely different applications, designed for completely different purposes. Topaz is a more specialized tool and it handles photo rescue as seen in the photos from this article, while Luminar is a combination with constantly added new.

Author: ekrzychu

Sharpening hair/fur with the same strength as the rest of the photo is a difficult task, and to achieve great results, it’s usually necessary to create a mask and sharpen the eyes strongly while sharpening the hair/fur much less. Here, doing everything at once didn’t produce stunning results, but it’s not bad either.

Author: Paweł

Below is the result of just increasing the resolution of a photo with two types of blur:

Author: David King

Noise reduction is also done automatically, and sharpness is improved, but there is no control over these parameters when upscaling. That is, there is a certain minimum that is automatically done and an be further enhanced, such as enabling sharpening. However, in the case of this photo, it didn’t work out because we have motorcycle motion that appears to be perfectly sharp, while the wheel could still be spinning. Anyway, the attempt to remove motion blur ended up like this in this case:

Author: David King

I also have another interesting example with double blur caused by depth of field and motion blur. Although the motion has been eliminated, the depth of field is still naturally distributed:

Photo: Błażej Zalesiński/studio dog photography – Inoinu psi fotograf

Photo AI has three sharpening settings: standard, lens, and motion, but almost always, the standard option does the best job. I also have an example of a photo that almost all of us will use Photo AI for:

Author: Bartłomiej Bałaban

Author: Bartłomiej Bałaban

I suspect that Photo AI will be used largely for souvenir photos, so ease of use is a great advantage here. Such tools should not be made just for professional photographers for large commercial projects – there are already three separate programs available for them.

Another example illustrates the progress that has been made within six months since the premiere of Topaz Photo AI. The change is significant enough that I updated the photo below. Now the result is such:

Photo: Błażej Zalesiński/studio inoinu (see profile on Instagram)

However, once I stepped out of the depth of field, the image came out jagged and when I tried to mask it, I couldn’t do it because the function was not working. Now it works and the area to be masked was much smaller than before.

I have been using Photo AI since its release

I decided to buy it immediately, and now, I can confidently say that these three functions: denoising, sharpening, and upscaling – work great together, because noise, poor sharpness, and low resolution are common in almost all photos taken when digital cameras were new. So, in addition to current photos, we can save a lot of old photos from the past. Moreover, photos come out blurry because, for example, it was dark, and the exposure time was extended, and high ISO is also used at such times. So, it’s good that we have these three functions together, as they are often used all at once.

Author: Zbigniew Skałka

In the example above, the hair suddenly and unrealistically goes beyond the depth of field. One of the functions of Topaz Photo AI is “face recovery”. This option extremely improves the sharpness of faces that have been out of focus, but the elements around them are still recovered in a standard way, so the depth of focus of the neck or, as in this case, hair may not always match.

Author: Zbigniew Skałka

Above, we have a very unusual and strong movement that could not be reduced.

The photo frames in this article are often not exactly the same like originals. I was approaching the main section, which was “rescued” by Topaz AI. Otherwise, in the web size, the result would not be clearly visible on wide shots.

Photo AI is also useful for enlarging graphics generated by artificial intelligence, as they are currently in very low resolution, and with Topaz, they can be enlarged few times, although for my internet purposes, a two-fold increase is usually sufficient.

The image generated in DALL-E 2 and upscaled using Topaz Photo AI.

Or something more sketch-like:

The image generated in DALL-E 2 and upscaled using Topaz Photo AI.

In practice, it looks like this: I upload a file to the program that is not suitable for publication, for example, because it is not sharp or it has too low resolution (or both), and then it is processed by algorithms and AI. I can then define how strong the noise removal, sharpening, or enlargement should be.

Graphics generated in DALL-E, processed in Topaz Photo AI.

Topaz Photo AI works both as a standalone application and as a plugin for Photoshop, Affinity Photo, etc. However, when upscaling an image, you cannot use the plugin, which makes sense because the entire document would have to be rescaled, not just the layer you are working on.

Graphics generated in DALL-E, processed in Topaz Photo AI – upscaling locked in Photoshop.

Theoretically, it is best to upload RAW files and Topaz also recommends this, however you then do not have control over the development process like in Lightroom. Later conversion to DNG using Photo AI is no longer a RAW with all the information.

The following example already made an impression during the premiere of Photo AI and looked like this:

The image generated in DALL-E 2 and upscaled using Topaz Photo AI.

However, the progress that has been made is huge and small gaps between objects/lines are now being significantly improved.

I asked readers for photos for this article and mainly JPGs came, and I also have old photographs of poor quality, mostly in JPG format.

Author: Paweł

The final effect is sometimes better, sometimes worse, for example in the following photo I didn’t manage to do the hair and shadow on the lip properly, but the fact that something came out of such a flawed sharpness makes an impression:

Face recovery, sharpening and noise reduction in Topaz Photo AI

This is also one of the examples that I regenerated on the current version of the program, because around the time of its release it couldn’t handle hair at all – some parts were completely blurred while others weren’t. Now it’s still like this, but the transition between these areas looks so much better.

However, let’s remember that it is AI generating results despite the lack of appropriate data. A reconstructed tooth may not look as good as on the “upgraded” photo and it also happened in the above example, where light reflection was misinterpreted, just like a shadow on a lip. In the case of completely out of focus people, their “recovered” appearance may differ from reality.

Author: Paweł

Enlarging photos in Photo AI is often as good as in the separate Gigapixel AI plugin, but the control over the final effect is more limited. Do you still remember the first photo?

I thought to myself, “I wonder if the version on the right could be further improved by putting them into Gigapixel“… Well, it turns out they can:

Topaz Photo AI

The enhanced original has a gigantic resolution, but I had to downsize it for the blog. Of course, there are visible defects, but I would like to remind you where I started and where I finished:

 

And going back to Photo AI… It’s time for a photo of a different kind. On the image below, I did noise reduction and sharpening:

Author: Robert Kozak

The same thing, only closer:

Author: Robert Kozak

The problems with the version I was using

Topaz releases updates every week, and the could improve the A.I. a lot. So the flaws I will mention may be outdated after a few weeks, but it is unpredictable – they may also persist for years.

The noise reduction works very strongly, even at the minimum setting. As a result, faces can look more plastic than if there was greater regulation. This is what noise reduction looks like in the “strong” version (there’s also a “normal” version for low levels of noise) set at 1% power:

Author: Przemysław Szklarski

In my opinion, this is an obvious mistake rather than a deliberate effect. Additionally, if noise reduction, upscaling etc. is not used, but the face recovery function is enabled, faces do not have noise even if the rest of the photo does – here a better AI training may be necessary rather than just a program correction. Currently, I simply run noise reduction and it’s done.

Author: Stanisław Dziura

Basically for just noise reduction, DeNoise run separately is better. However, my second biggest concern is already fixed, which means the masking works.

Author: Przemysław Szklarski

Hardware requirements and license

Author: Paweł

A short video showing Photo AI in action (just in Polish)

I recommend starting watching from 00:01:28 because earlier I inform viewers about this article and you already know about it :).

 

Update: Strong noise reduction now uses a new AI model and is even more effective, many smaller improvements have also been introduced.