Reducing noise of high-ISO Milky Way photo with Topaz Photo AI V2

Reducing noise of high-ISO Milky Way photo with Topaz Photo AI V2

How does the noise reduction work on the new Topaz Photo AI V2? We kicked the tires on its noise removal feature.

Milky Way photo, Mammoth Lakes CA.

What is Photo AI?

Photo AI is an image enhancement app that sharpens, reduces noise, and increases the resolution of your photos through machine learning. It works on either Mac or PC. And it also works as stand-alone software or in Lightroom Classic or Photoshop CC. It also supports RAW files, as well as JPGs, TIFFs, and DNGs. It allows you to refine your selection using masks, preserve text in photos, recover the detail in faces and much more. And finally, it supports batch processing. 

The price is currently $199 and is not subscription-based. You will most likely be able to use the app for many years. Why do I say this? I purchased Topaz Star Effects in November 2013, initially using it with Photoshop CS6. Topaz Labs hasn’t made Star Effects in years. Despite that, it still works in Photoshop CC. Topaz also transferred it to my new email when I lost access to my previous email. Topaz, take a bow for creating apps with such longevity and providing solid customer service. This should never be overlooked.

What is the general workflow of Photo AI?

When you load a photo, Photo AI will examine for numerous parameters. When loading RAW files, for example, Photo AI will correct lens distortion by reading the RAW data. Then it will attempt to detect the subject, faces, and noise and blur levels. From this, it will calculate auto-suggested values. Of course, you have the opportunity to change the parameters later via a very simple, well-laid out interface.

Using the noise removal tool in Photo AI

  • Very good and surprisingly quick denoising
  • Supports RAW files as well as JPGs, TIFFs, and DNGs
  • Sharpening (deblurring) in noise removal tool works very well
  • Extremely easy to use
  • Clean interface
  • Autopilot works well at identifying what you might want to do upon opening photo
  • Clean, simple and logical user interface
  • Excellent camera and lens support
  • Works as stand-alone or with Lightroom Classic or Photoshop
  • Stable and reasonably quick

Okay, that’s all fine. But is the noise removal tool any good?

Yes. Okay. I’m done here. That’s the end of the article.

Just kidding. 

Being a night photographer, I had to see how it did with complex noise from a high-ISO Milky Way photo. To do this, I used it within Lightroom Classic and as a stand-alone app.

Removing noise within Lightroom Classic

1711472982 813 Reducing noise of high ISO Milky Way photo with Topaz Photo | Theedgesm1711472982 887 Reducing noise of high ISO Milky Way photo with Topaz Photo | Theedgesm

Left: Original file. Right: Noise reduction applied with Photo AI. Settings: Normal AI model, Strength 75, Minor Deblur 50, Original Detail 100.

When you use Photo AI within Lightroom Classic, it automatically creates another TIFF file, as is the case when using many other applications within Lightroom.

Here are some screenshots. I tried three different settings in total while in Lightroom Classic.

Photo AI V2 within Lightroom Classic with TIFF file at 75 Strength using “Normal” AI Model

Screenshot of Photo AI Remove Noise. Left: original. Right: Remove Noise applied.
Screenshot of Photo AI Remove Noise. Left: Original. Right: Remove Noise applied. 100% view.
Screenshot of Photo AI Remove Noise. Left: original. Right: Remove Noise applied on a stronger setting.
Screenshot of Photo AI Remove Noise. Left: Original. Right: Remove Noise applied. 200% view.

Photo AI V2 within Lightroom Classic with TIFF file at 48 Strength with “Strong” AI Model

Screenshot of Photo AI Remove Noise. Left: original. Right: Remove Noise applied using the Strong" AI Model.
Screenshot of Photo AI Remove Noise. Left: Original. Right: Remove Noise applied using the Strong” AI Model. 100% At this point, the image is losing detail and is starting to look a bit “plastic.”
Screenshot of Photo AI Remove Noise. Left: original. Right: Remove Noise applied using the Strong" AI Model. 200% view.
Screenshot of Photo AI Remove Noise. Left: Original. Right: Remove Noise applied using the Strong” AI Model. 200% view.

Using the noise removal tool as a stand-alone app with RAW files

As previously mentioned, Topaz Photo AI works as a stand-alone app. When used like this, it can process RAW files from almost any camera. Would removing noise be that different?

I processed the same photo. Because we were no longer in Lightroom, my light processing of the image is not visible.

Loading the RAW image into Photo AI, it again examined the photo, it again attempted to detect the subject, faces, and noise and blur levels. From this, it used Autopilot to calculate auto-suggested values. The next two screenshots below are a comparison between the unprocessed photo and one with reduced noise using auto-suggested values. By the way, I loaded the photo three times in a row, and it consistently gave me the exact parameters.

Photo AI V2 with RAW file at Default Strength

Left: original. Right: Remove Noise applied using the "Normal" AI Model on a RAW file. 100% view.
Left: Original. Right: Remove Noise applied using the “RAW Normal” AI Model on a RAW file. Default. 100% view.
Left: original. Right: Remove Noise applied using the "Normal" AI Model on a RAW file. 200% view.
Left: Original. Right: Remove Noise applied using the “RAW Normal” AI Model on a RAW file. Default. 200% view.

For fun, I cranked up the Strength Slider to 50. You can see that it’s starting to lose some detail, although I certainly could turn up the Minor Deblur (a sharpener) to see what it does. You can see these below.

Photo AI V2 with RAW file at 50 Strength

Left: original. Right: Remove Noise applied using the "Normal" AI Model on a RAW file. Strength turned up to 50. 100% view.
Left: Original. Right: Remove Noise applied using the “RAW Normal” AI Model on a RAW file. Strength turned up to 50. 100% view.
Left: original. Right: Remove Noise applied using the "Normal" AI Model on a RAW file. Strength turned up to 50. 200% view.
Left: Original. Right: Remove Noise applied using the “RAW Normal” AI Model on a RAW file. Strength turned up to 50. 200% view.

When using Photo AI within Lightroom, it was acting on a TIFF file. I could push it pretty hard, and it would keep reducing noise, although it would begin looking increasingly “plasticky” and fake.

What would happen if I boosted the RAW file when it was using different algorithms? The results were fascinating.

Photo AI V2 with RAW file at 60 Strength

Left: original. Right: Remove Noise applied using the "Normal" AI Model on a RAW file. Strength turned up to 60. 100% view. Notice the artifacts, particularly by the Milky Way.
Left: Original. Right: Remove Noise applied using the ” RAW Normal” AI Model on a RAW file. Strength turned up to 60. 100% view. Notice the artifacts, particularly by the Milky Way.
Left: original. Right: Remove Noise applied using the "Normal" AI Model on a RAW file. Strength turned up to 60. 200% view. Notice the artifacts in the top right of the image.
Left: Original. Right: Remove Noise applied using the “Normal” AI Model on a RAW file. Strength turned up to 60. 200% view. Notice the artifacts in the top right of the image.

Switching to the “RAW Strong” AI Model yielded similar results. Of course, Topaz’ Autopilot default had been just fine. It’s just fun to see what happens when you push things farther along.

Photo AI V2 with RAW file at 60 Strength using the RAW Strong AI Model

Left: original. Right: Remove Noise applied using the "RAW Strong" AI Model on a RAW file. Strength turned up to 50. 100% view. The artifacts are very present.
Left: Original. Right: Remove Noise applied using the “RAW Strong” AI Model on a RAW file. Strength turned up to 50. 100% view. The artifacts are very present.
Left: original. Right: Remove Noise applied using the "RAW Strong" AI Model on a RAW file. Strength turned up to 50. 200% view. The artifacts are very present.
Left: Original. Right: Remove Noise applied using the “RAW Strong” AI Model on a RAW file. Strength turned up to 50. 200% view. The artifacts are very present.

It was interesting to see that with the TIFF files, you could push the image farther without incurring blatant artifacts. It would instead grow increasingly smooth and “plasticky.”

The program seemed to handle color noise quite well without affecting the Milky Way.

How similar is it to Topaz Denoise AI?

Topaz Denoise AI is a denoising software that is no longer available. How similar is this to the denoising software in Photo AI V2?

It does not quite have the same interface as Topaz Denoise AI although if you are comfortable using that, you will have no issues with Photo AI. In fact, instead of four AI models, you get two or three, depending on what image file you are using at the time. You also don’t have individual controls over color noise like you did in Denoise AI. The interface is much simpler and there are less parameters. However, as long as it doesn’t destroy the stars or add artifacts, it’s all good. And Photo AI V2 does all that.

I asked Topaz about the similarities between the programs.

The RAW models and the base noise removal models in Photo AI are unique compared to the ones in DeNoise AI, as they have been trained on different data and produce superior results in most circumstances. The “Extreme” denoising model is an improved version of the “Severe” denoise model which appeared in DeNoise AI. Overall the models are similar in speed, but with generally improved quality of results.

Photo AI also has improved RAW and color handling capabilities which were not present in DeNoise AI, which can make a big difference. The controls are also improved in Photo AI, so the sliders for strength and debtor as well as recovering detail have a stronger effect on the image results as compared to DeNoise.

Lastly, in the latest version of Photo AI, users have the ability to not only stack multiple denoise models but also combine them with other filters and selections in order to achieve results which were not possible in DeNoise AI. For example, de-noising only the background using a strong model, and applying a light denoise effect to the subject of the photo. Likewise, even in photos with extremely high noise, users can stack multiple denoise effects on top of each other as needed.

Topaz Labs product team

The ability to stack denoise models was not present in Denoise AI, as they mention. However, I believe if you are using Denoise AI in Photoshop, you can get around this by using the various tools such as Layers, Layer Masks, luminosity masks, and similar features.

More musings about Photo AI V2’s noise reduction

So far, the noise removal tool works well on a complicated high-ISO Milky Way image. Also, the program is very easy to use. Furthermore, at least at reasonable settings, it doesn’t “damage” the quality of the stars or galactic core of the Milky Way. It does, however, get rid of color noise and high-ISO noise quite well.

Although I often stack my Milky Way photos to reduce noise, I have found that with more modern noise reduction tools such as the very capable one found in Photo AI V2, I feel increasingly comfortable taking single-exposure high-ISO photos.

Offering the capability to work on RAW files as well as within popular platforms such as Photoshop or Lightroom Classic is also a welcome improvement. This enables most of us to incorporate Photo AI V2 into our workflow.

Lastly, the controls in Photo AI V2 are simpler than Denoise AI. I find it interesting that this — and Lightroom’s denoise AI, for that matter — have extremely simple controls while producing extremely good results. We’ve seen almost hockey stick-like growth from machine learning programs, as evidenced in part by how good AI-generated art has gotten. More effective noise reduction reflects this, as new models targeted at removing noise become increasingly accurate at identifying unwanted elements. At least in this case, this is where AI can help us with the photo that came from our camera. Certainly, Photo AI V2 does that extremely well, even with a complex high-ISO Milky Way photo with lots of stars.