through the F-Mount -  photography by Jürgen Becker
home galleries articles links about contact

Topaz Labs Gigapixel AI 4.0 - artificial intelligence for digitising!

April 6, 2019

More and more, our daily life is surrounded by AI helpers. Photography is no exception - think of the face detection in your camera for example. Gigapixel AI from Topaz Labs is a stand-alone postprocessing tool that offers AI-driven image upsampling. Here, AI-driven means the additional pixels are not just interpolated, they are set by the neural network of Gigapixel AI that has learned its job through thousands of real-world images!

If your image doesn't contain enough pixels for printing/viewing it in the size you want, Topaz Labs recommends the usage of Gigapixel AI. For me, that is a very rare case. However, I found Gigapixel to be a great and easy-to-use tool for improving digitised images.

Below, you can see a typical digitised image of mine. It shows the little fishing village of Siglufjörður in northern Iceland. In June 1989 I captured this image on ISO 100 Fujichrome RD slide film. 12 years later, I scanned the slide with my Nikon Coolscan III. Such images have about 3600x2400 pixels, due to the optical resolution of my scanner of a bit more than 100 pixels per mm. That is enough pixels for me, but in fact the image is only usable for prints up to about 30x20cm. If you view this image fullscreen on a typical 24" 1920x1200 monitor, it looks a bit soft all over. The output of your DSLR contains more details and is sharper!

Below is a 100% crop of the above image:

Topaz Labs has just introduced version 4.0 of Gigapixel AI. I downloaded a trial version and tried to improve the image above. Gigapixel allows a scaling of 0.5x, 2x, 4x or 6x. Moreover there is an option for custom scaling in the range from 0.5x to 6x. My first approach was to use a custom scaling of 1x. Gigapixel just needed one second for processing. But the result disappointed me - it was just a slight improvement, like a standard sharpening. Next try: I set Gigapixel to 2x. Now, processing took around 10 minutes! Then I compared the result at 50% with the original at 100% - again, just minor improvements! Last try: I downpixelled the original image by a factor of 2 in Photoshop Elements! Then I put Gigapixel into service with an upscaling of 2x. The processing time was around six minutes. What can I say - the result has blown me away! Compare the 100% crops below for yourself:

Of course, I tried many digitised images of different ages and different subjects - see some examples below. Mostly, the results were great. But there is one restriction: for very noisy images the output of Gigapixel doesn't really look good. Examples are most of my slides from the 80s on ISO 400 Fujichrome RH film.

I bought a regular version of Gigapixel AI and I am happy that a lot of my slides are really back in the game now!


The results of traditional unsharp masking are not comparable with the results of Gigapixel AI! I also downloaded a trial of Topaz Labs' Sharpen AI. That's a powerful tool, too. But I didn't get results as good as those from Gigapixel AI. Moreover, the processing times are even longer. I will wait for the next version of Sharpen AI...

How to use Gigapixel AI?

Gigapixel AI only has very few settings options. The screenshot on the right shows the ones I used for the image of Siglufjörður. In my opinion the "Remove Blur" setting defines a kind of threshold value for blur. If the blur is stronger, Gigapixel will not sharpen it, else it will! If the original piece of film was curved or wavy (a typical problem for digitising), you might need the highest setting! Simply try! The "Suppress Noise" setting is very effective. It is a much better alternative than a seperate noise reduction after Gigapixel processing! For scaling I have only used the 2x setting, yet.

@Topaz Labs: more AI, please!

What I would like for my digital photography processing is a tool I would name "Bokeh AI". Many of my wildlife images contain parts with not so nice bokeh. Especially the transition zones often are a bit nervous. I think that can also be solved with artificial intelligence. Maybe, it would be a good idea to combine Bokeh AI and Sharpen AI.

More examples:

Please note: every Gigapixel image shown below was created with the above-mentioned method, meaning downscaling by 2 with Photoshop Elements and upscaling by 2 with Gigapixel AI 4.0!

View of Ravello, a little village in Southern Italy. Captured by my Father in March 1958.
The ISO 50 Agfa CT 18 slide film was very curvy after having been 60 years in a sealed slide frame. Therefore, in terms of sharpness the digitised image is poor in many parts of the frame. Below you will find 100% crops, on the left from the original image. The middle image was processed by Gigapixel AI with "Remove Blur" set to None. On the right it was set to "High" (which is the maximum setting):

Here are two 100% crops of another image from March 1958 - the offerings of a fruit dealer in Naples, Italy. My father used a flash here - the slide film was Agfa CT 18 again. Firstly, the crop from the original image:

Now the same crop from the Gigapixel output:

Back to Iceland, June 1989, 100 ISO Fujichrome RD slide film:
Breeding kittiwakes in the cliffs of Arnarstapi, 100% crop, from the original and then from the Gigapixel result:

The next slide was captured in Lofoten/Norway in September 1990. It shows a goat eating seaweed at low tide.
Below two 100% crops, the left one from the original, the right one from the Gigapixel processed image:

At last a slide from May 2001, showing my younger daughter. It was captured on ISO 100 Fujichrome Sensia slide film:

Here are 100% crops of the above image, the original to the left, the Gigapixel AI output to the right:

© 2019  All rights reserved.