Topaz Labs Gigapixel AI Software v5.8.0

This review is of Topaz Labs Gigapixel AI version 5.8.0 using the GPU as the AI processor.

The Topaz Labs Gigapixel AI software promises to increase resolution not just simply enlarging pixels. Examples include:

Add image detail that looks completely natural
Enhance the texture of feathers
Improve skin texture and sharpness
Higher-resolution landscapes
More detailed cityscapes
Crystal-clear upscaling with no blocky artifacts
Fix extreme pixellation in low-resolution images

Some improvements in this release are quality improvements using a GPU in Windows (which is the case here) and reduced memory usage.

Here is an example from the Topaz Labs website. After scouring the Internet, the source of the below image was found in full resolution so it can be compared.

Let's run the first image on this page through Gigapixel AI v5.8.0 and scale it 2× to see if similar results can be had.

Though there are differences, just like with Topaz Labs Gigapixel AI v5.7.3, none of the above images look as good as what is promised by the Topaz Labs website. Here is what the website promises again.

Now, let's downsize to 450×400px and use high-compression with the original low-resolution example from the website and then try again.

The result appears like a joke when compared to the Topaz Labs website's examples.

Perhaps it is unfair to use the examples from the website even though this seems perfectly reasonable. Let's use the original source to see if we can acheive the results that the Topaz Labs website shows.

Here, the original source is compressed with high-compression.

Here, the original source is compressed with medium-compression.

Here, the original source is compressed with low-compression so as to retain as much detail, like the bricks and mountains, as possible.

When compared to Topaz Labs Gigapixel AI v5.7.3, the new algorithms in v3.8.0 appear to create sharper images.

Conclusion

This software is no where near as effective as the Topaz Labs website makes it look, but it is an improvement over the last version.

If there appears to be flaws in this testing methodology then please inform.