Topaz Labs Sharpen AI Software v4.0.0

This review is of Topaz Labs Sharpen AI version 4.0.0 using the GPU as the AI processor.

The Topaz Labs Sharpen AI software helps to rescue blurry images that would normally be thrown out by sharpening them with artificial intelligence. Here are some of their promises:

Reduce perceived movement
Focus on the eyes
Recover image detail
Increase real (not simulated) sharpness
Rescue multiple types of motion blur
Save misfocused images

Using with Noise Reduction Software

In reviewing the Topaz Labs DeNoise AI v3.5.0 software, some very soft images resulted. The noisier the original image, the softer the resulting cleaned image was. Can Sharpen AI v4.0.0 restore/create detail that is not there?

Here is the source image that had the noise reduced. Notice that the image detail almost looks like a painting due to its lack of detail.

After applying a "Too Soft - Very Blurry" sharpen model to the image, this was the result. Notice that the AI introduced black hair into the dog's coat. Those are not really there. This is the dark colored dog jacket behind the hair. To be fair, it may be the DeNoise AI that confused Sharpen AI here. There are also black hairs on the dog's head and tail. This it not really different from v3.3.5 of the software.

Blurry Image

Here is an image of a photograph taken in the dark so as to get a realistically blurry image.

The sharpen model "Standard" was used with the "Automatically detect recommended settings" turned on.
The sharpen model "Motion Blur - Normal" was used.
The sharpen model "Motion Blur - Very Blurry" was used. It is an improvement.
The Photoshop sharpening tools were used to sharpen the image.

It's not perfect, but it is better than using Photoshop's sharpening tool. The claimed improvements in v4.0.0 come to the Standard model, but the results from this model do not look natural at all. Probably later versions of this software will have improvements made to this model since it is all new. It is recommended to turn down the sharpen slider from what "Automatically detect recommended settings" suggests if using the Standard model until improvements are made.