Topaz Labs DeNoise AI Software v3.6.1
This review is of Topaz Labs DeNoise AI version 3.6.1 using the GPU as the AI processor.
At the time of writing this (2022) the Topaz Labs website demonstrated removing noise and recovering detail that was lost with remarkable results. Let's put Topaz Labs DeNoise AI to the test and see how well it really does.
Here is an example from their website.
This software works with both JPEG and RAW formatted image files. Best results can be had using RAW images.
This software package includes a photoshop filter plug-in.
Modest Noise Example
To cure modest noise is probably the most common use of this software.
To test, RAW images of a dog will be used since there is a lot of detail lost in the fur due to the noise. First, a modest ISO 7200 image from a Nikon Z7 will be used.
Here, the "clear" option was used. The "high" noise setting seems like too much but it really is up to personal perference. There appears to be absolutely no difference from v3.5.0.
Here, the "low-light" option was used.
Here, the "RAW" option was used. It requires a RAW image file.
Here, the "severe noise" option was used. It may be designed for more noise than is in this image.
Here, the "standard" option was used.
For comparison, here is the original shot at ISO 64 (no noise).
Severe Noise Example
Here, a ISO 25600 image file is used to test with. This ISO pushes the limits of the Nikon Z7.
DeNoise AI v3.6.1 is not as sharp as v3.5.0. Use Topaz Labs Sharpen AI v4.1.0 to sharpen this image, and perhaps this is the plan of Topaz Labs.
Conclusion
Except in RAW, the differences are not very evident, but there were some differences in the saved file size. The new algorithms created files that could be better compressed which means more noise was removed (and possibly less detail).
Topaz Labs DeNoise AI does live up to its name. It does a good job "denoising" while preserving some detail which is something Adobe Photoshop cannot do well. For the most part, the improvements are subtle compared to v3.5.0, but they are there.
Related links: