Night photographers are always looking for good noise reduction software. After all, noise is the enemy of night photography. Extremely low light coupled with long exposures and high ISOs can make for rather noisy photos. How would Topaz DeNoise AI handle this?

Radio telescopes looking for signs of … artificial intelligence? Have they found it in Topaz Labs DeNoise AI?

The different noise reduction modes

I downloaded the latest version of Topaz DeNoise AI, which was version 3.2.0. With version 3, Topaz states that they have created a completely new AI engine and offers a new low-light model to give better details in dark areas. Also new is Severe Noise mode.

They also renamed some of their other modes:

  • Standard mode was formerly Denoise AI. This is what Topaz describes as a well-rounded noise reduction coupled with detail preservation.
  • Clear was formerly AI Clear. This is for denoising smooth surfaces
  • Severe Noise is, as the name implies, for addressing excessive noise from high ISO.
  • The new Low Light mode is intended for challenging lighting conditions with lots of noise.

Selecting a low-light high-ISO image for testing

I chose to test DeNoise AI with a night photo of some radio telescopes I had shot at ISO 4000 for 20 seconds. Often with this sort of photo, I would photograph the foreground using a longer exposure, low ISO setting and blend that clean version with the sky. However, it was 2:30 a.m., and I forgot. It happens. Consequently, I had a lot of noise throughout the image.

Would DeNoise AI be able to denoise the foreground well? Would it keep fine details intact? And would it dull the sharpness of the stars? A tall order indeed.

Testing in Photoshop

Partially out of habit, I do a lot of my processing in Photoshop. Thankfully, DeNoise AI works as a plugin in Photoshop as well as Lightroom Classic. 

I immediately took a liking to the comparison view as well as the ability to quickly switch between different views. 

software, Topaz, Denoise, night, noise
Comparison View in Denoise AI. In this example, if you wish to see Clear View, you would select that at the right. Click on the image for a larger view.

I set all the mode settings to the same, just for the sake of comparison. To me, the Low Light mode looked the most pleasing. It kept the sharpness of the stars intact. And it denoised the radio telescope. 

Zooming in close for some pixel peeping

After applying the effect to a new layer in Photoshop, I wanted to compare before and after more carefully. I zoomed in to 200% to further compare. Nobody except the person doing the processing ever looks at photos this closely. And no one would see this level of detail on any website, and certainly not on a phone. Nonetheless, I felt this would reveal more about how Denoise AI did.

Above, you can see a zoomed-in, close-up 200% view of the radio telescope and the stars. Toggle back and forth to view “before” and after”  images.

Shown above is a zoomed-in, close-up 200% view of the supports for the radio telescope, which were very dark and had considerable background and high ISO noise, partially from being dark.

As you compare the “before” and “after” images, you can see that DeNoise AI has fixed the noise quite well while keeping details intact. It even kept the faint stars sharp.

One word: Wow.

I could end my conclusion with one word: Wow. 

I’ll elaborate further. I’m deeply impressed with how well DeNoise AI is able to reduce noise while preserving details. Their AI engine is impressive at differentiating what is noise and what is detail that should be kept. It also preserves contrast beautifully. I have never seen any other noise reduction software that can effectively do both simultaneously.

To preserve this amount of detail, I’ve previously needed to fine-tune the adjustments using a combination of luminosity masks and “painting” on the layer mask to add or remove the noise reduction. It also has an excellent interface and is exceptionally easy to use.

Are there any cons? Sort of. The software requires slightly higher computer graphic card requirements. And sometimes, the processing time can seem slow compared to other software.

So why do I say “sort of”? It comes back to its quality and preservation of detail while denoising. Sure, it takes DeNoise AI a few seconds to provide previews in comparison view. And sure, it may sometimes take as long as 40 seconds to render after applying the effects. I would submit, however, that DeNoise AI is reasonably fast when you consider the processing involved.

Further, DeNoise AI is considerably faster and less work than it would take me to fine-tune the adjustments. That’s something I’ll happily embrace every single time.

Topaz DeNoise AI is the best noise reduction software I have ever used. And coming from a night photographer, that is really saying something.