Incredible new AI noise reduction in LR / ACR

Adobe Camera RAW (ACR v15.3) and Lightroom (Classic v12.3 / Desktop v6.3) have just added a powerful new noise reduction tool using artificial intelligence (as well as some great enhancements to HDR capabilities and more noted at the bottom of this post). Noise reduction is a powerful tool not only for high ISO images, but also noisy images from small sensors (such as drones) or helping to make larger prints from low ISO images. There are now a number of AI-based tools and ACR is already among the best in its first release. In this tutorial, you’ll learn why you should use it, how to get optimal results, and how it compares to DXO PureRAW.

In my experience, this tool consistently adds a lot of value for its intended targets: noisy images. It’s is extremely helpful for noise in shadows, high ISO images, and small sensors (such as images captured with a drone). It can help make larger prints from images captured at an optimal ISO. And it can even reduce hot pixel noise in some images. The final result is significantly less luminance noise, less color noise, and better preservation of detail compared to the older manual noise removal.

 

Workflow for Adobe Denoise:

  1. Open or select an image in LR or directly in ACR (the feature is not available in inside RAW Smart Objects and the RAW Filter does not actually work with RAW data). You may also <shift>-click to select multiple images at the same time.
  2. Consider using exposure or shadow adjustments so you can clearly see any noisy areas you’ll want to review in the next steps (this won’t affect the results and you can change/undo it later).
  3. In the Detail tab, click the “Denoise” button. If you <alt/option>-click Denoise, it will run “headless” and immediately process the image with the same settings used last time you ran denoise.
  4. The preview shows 100%. You cannot change the size of the preview, but you can easily pick other parts of the image to preview by clicking directly on the image. You may alternatively click the – icon (or <alt/option>-click in the preview) to zoom out, then click elsewhere in the preview to zoom back in.
  5. Click the preview to view before/after to help choose the desired amount of Denoise. 50% is generally a good amount.
  6. Click “Enhance” then you may open the new image which shows in the film strip. You may wish to <ctrl/cmd>-click to select the original as well so that you may blend the two.
  7. You may apply manual noise reduction in addition to Denoise. That’s not something you should do often, but it can be useful in some cases.

Some tips for working with Denoise:

  • Expect to make some minor changes to adapt your approach if you’re already using another noise reduction tool. Adobe Denoise is designed to reduce noise. It does add some detail, but not  real sharpening like DXO or Topaz DeNoise do. So you’ll need to add some degree of sharpening or detail enhancement to the Adobe Denoise image if you’re hoping to match the level of detail those products tend to produce by default.
  • You can apply Denoise at any time on the RAW file, but it should ideally be done before using any tools which make permanent changes based on the current pixels (this includes AI-based masks like Select Sky/Subject and the healing brush). Denoise will automatically update those areas, but it’s a good idea to review them if you denoise after those changes. I’ve also seen denoise make some slight shifts in apparent color/tone. They’re quite minor, but it’s a good idea to look for changes if you’re denoising an image you’ve already processed.
  • You can skip the popup interface and run with the last same settings used last time by <alt/option>-clicking Denoise (the … will disappear on the button when you’re holding the correct shortcut key and hovering).
  • Adobe Denoise also adds detail, which can be both a benefit and a potential concern, depending on the image content. automatically turns on the “RAW Details” enhance option. If you compare Denoise at 1% (where it’s doing almost nothing) vs unchecking Denoise (so that it’s completely off), there is a significant change in detail. In other words, it adds a lot more detail beyond that provided when you only have “RAW Details” enabled. This extra edge detail increases with higher amounts of Denoise. This has several implications:
    • You may turn on Denoise at a very low percent to help reveal more detail in an image (like a form of AI capture sharpening).
    • You may find this detail results in artifacts in some areas. So even if you’re already comfortable with how  RAW Details affect your images, you should review the results closely since they’re different now. Watch out in particular for halos along strong edges like backlit buildings. If you run problems, you can blend locally with the original, blend with a version using less Denoise, or just use the old manual noise reduction as needed. It’s not a concern I’ve seen in many images, but you should be aware of it. It’s also the sort of problem I expect may be eliminated as this tool matures with future updates.
  • One scenario that I find Adobe Denoise doesn’t handle well yet: very high contrast edges, such as a sunset sky behind the hard edge of a building. In that scenario, you may see halos. Hopefully this is addressed in a future update, but there may be a few scenarios where another approach is preferable or should be blended in via layer mask.
  • If you wish to filter LR to only show denoised images, you may search on the text “enhanced” and may further limit metadata to file type of “digital negative / lossless” to show DNGs (in case you have unrelated files with a similar name). You may also go to Settings / File Handling and check “automatically add keywords to enhanced images”, which will cause “Denoise” to be added as a keyword (the amount of noise reduction is not noted in either the keyword or new file name).
  • There’s no direct way to to filter to files which have not been denoised, but you could use a creative approach by setting a metadata filter for file type to “Raw” and opting for Denoise to output as a stack. The resulting DNG will be at the top of the stack and will not show in the filtered stack (so long as you leave the stack collapsed). This assumes you did not import your images as DNG, in which case both the source and denoised image would have the same “file type”.
  • Speed with this tool varies wildly based on your computer. On my M2 Max, converting 10 D850 images took an average of 27 seconds per image. I’ve heard reports of much longer times with much older computers, so your speed will depend heavily on your computer’s capabilities. If you have a slower computer, I recommend just letting the batch run in the background (you can even keep working on other images in LR if you like). Adobe’s official guidance is: “For best performance, use a GPU with a large amount of memory, ideally at least 8 GB. On macOS, prefer an Apple silicon machine with lots of memory. On Windows, use GPUs with ML acceleration hardware, such as NVIDIA RTX with TensorCores. A faster GPU means faster results.”
  • You can get basic info on your GPU in PS under Help / GPU Compatibility. If you’d like to compare with others, some GPU benchmark options which have been recommended to me are 3DMark (for PC) and Cinebench for (PC or MacOS).

 

Adobe Denoise vs DXO PureRAW 3:

There are a few AI-based noise reduction tools out there and I’ve posted tutorials previously on DXO PureRAW 3 (with “DeepPRIME XD”). How does ACR compare? The short answer is that they have complimentary strengths, so I prefer to use a mix of both depending on the image. The full answer is a bit longer as they aren’t fully comparable, as DXO targets a larger range of RAW enhancements.

Pros for Adobe Denoise:

  • Included with ACR and therefore costs nothing if you have Photoshop / Lightroom CC (vs $129 for PureRAW for a new purchase at full price).
  • Offers control over the degree of noise reduction, which can be helpful to fine tune the balance between noise reduction and preserving detail.
  • Less prone to artifacts in fine details.
  • Preserves the mosaic data. This may facilitate use of improved demosaicing algorithms in the future (which may help improve fine detail or reduce pixel-level artifacts). However, if you keep the original RAW, you’d always have this data (and the combined size of the original and DXO DNG is only about 5% larger than the Adobe DNG). So this is nearly a wash if you’re willing to keep the original when using DXO.
  • Simpler to use. There’s not much to think about here, which is nice. That said, the DXO interface isn’t too complicated.
  • Embeds fast load data, which may provide some performance boost when changing images in LR / ACR.
  • I find that it does a somewhat better job with high ISO night sky images. DXO tends to shows some artifacts (faux star trails) and makes secondary stars too strong (which makes for a cluttered star field where everything is a bright star). Adobe Denoise also makes secondary stars too strong, but the overall result is a bit better.
    • One benefit of this cleaner result is that you can combine Adobe Denoise with stacking multiple images. This would help achieve greater total noise reduction and/or allow you to shoot fewer images. I’d happily stack say 5 images instead of 10-20. Not only will that save time (on a typically cold night), but helps reduce problems with sky area near edges (where you may not have much data from other images in the stack).
    • I find that DXO does a better job with foregrounds, so I expect I’ll use both tools on the same image and use the sky from ACR and the foreground from DXO.
  • Direct integration with LR / ACR.
  • It’s a first release and only going to get better. Adobe’s Eric Chan noted in his blog post that they’re continuing to work on better training data, support for using Denoise with Super Resolution, and eliminating the creation of a new DNG file. If you look at the history of another new feature, HDR, you’ll see it has improved significantly in the past six months already (including the improvements noted below).

Pros for DXO PureRAW:

  • Does a better job of enhancing high ISO shadow details. It also includes a “lens softness” control to help control the degree of detail enhancement. When you want to make enlargements for print, I find DXO (especially when combined with Topaz Gigapixel) is an indispensable tool. This is a great reason to consider adding DXO to your toolkit.
  • 25% smaller files. I assume this is because PureRAW is saving demosaiced data (RGB) vs Adobe Denoise which preserves mosaic data (RGGB). For reference, a typical D850 is roughly 51MB for the original NEF, 135MB for the PureRAW DNG, and 178MB for the Adobe DNG. You might save the file space or use this as an opportunity to keep the original RAW for reprocessing in the future as these algorithms continue to improve.
  • Offers vignetting, chromatic aberration, and lens corrections. I don’t generally find these to be huge advantages. The lens corrections are very helpful if you get perfect results, but if you wish to blend in some of the original image, you need to skip them to align the image. I wouldn’t say I’ve found the chromatic aberration and vignetting addresses problems I can’t address with other tools in ACR.
  • Outputs to a sub-folder, which may be preferable in organizing the derivative DNGs.

Note that camera support varies a little bit here and both are likely to offer expanded options over time. I do not know if Adobe Denoise supports more cameras than PureRAW in general or just different cameras (you can check your images with DXO’s free trial to confirm with your own images).While both were able to process images from a DJI Mavic Pro drone, the results from PureRAW were not as expected and it does not seem properly supported as of v3.1.0.

Adobe Denoise is also able to process a RAW (not ProRAW) photo from iPhone 14, whereas PureRAW 3 could not. Note that the iPhone can capture two different types of RAW images and only one of them can be processed with Adobe Denoise. The native iPhone app RAW files are ProRAW files which are already demosaiced (partially processed), and therefore not compatible with AI Noise. You’ll have to use an app like ProCamera to capture in the mosaiced RAW format if you wish to use this software. You can tell which is which in ACR / LR based on Denoise availability, as well as by reviewing Metadata/DNG in LR and checking to see that “mosaic data” shows “yes”. In a brief test, I felt that the RAW + Adobe Denoise version showed more noise and less sharpening artifact, so it may be worth exploring this if you do serious photography with an iPhone (but the differences probably aren’t worth the effort for casual use).

I find these tools are very complimentary and I’m glad to have both. When to use which tool:

  • When you want noise reduction with the most natural look: ACR Denoise
  • When you want to enhance fine detail or restore very noisy shadow detail: PureRAW3
  • High contrast edges (such as sunset behind buildings): PureRAW3 (note that you may see better results using the option to remove chromatic aberration in ACR than in PureRAW3 in this scenario)
  • Starry night skies: Adobe Denoise or manual noise removal in ACR coupled with stacking. Adobe Denoise makes minor stars more prominent (which can make the sky too cluttered). You can get a better result by using a more modest amount of Adobe denoise and then add manual noise reduction. I also find Adobe Denoise may require a slight shift in tint to keep consistent color in the night sky. I’d like to see both tools improve their ability to handle starry skies.

This is just the first release of the tool, so the last couple of items may shift in favor of Adobe Denoise as ACR is updated over time. Ultimately, Adobe is coming out strong in their first release and I expect many photographers will decide it’s good enough (given it provides great results at no extra cost). However, I think PureRAW is excellent and there are compelling advantages for many photographers. I recommend trying DXO’s free trial to see for yourself, your degree of benefit will depend on the kind of images you capture.

 

Adobe Denoise vs Topaz Denoise:

Many people love Topaz Denoise. They make amazing products and I’m an enormous fan of Topaz Gigapixel. Topaz were early pioneers with AI noise reduction and have made a great product. However, as often happens, there’s lot of competing innovation and in my opinion leadership has now shifted to Adobe and DXO. In particular, I believe the RAW processing workflow is simply better with those products. However, if you don’t care about reducing noise at the RAW stage, then you’ll likely have a stronger preference for Topaz.

Pros for Adobe AI Denose:

  • You can apply the noise reduction at any time and trust that the new RAW (linear DNG) will look extremely close to where you started. That makes it very easy to migrate existing edits. that’s not the case with many files I’ve tested with Topaz. I see huge shifts in color balance, tonality, and sometimes overall vignetting of the image. It’s a different result, and in my experience complicates editing.
  • I prefer the results from Adobe, I find they have the least amount of artifact of any of the AI tools so far. But it depends on your image content, and there is no universally better tool.
  • It’s simpler, just a single Denoise slider. Topaz also has a reasonably simple interface, but you have to choose which model to use and set up to four sliders.
  • Topaz is not showing the RAW image as you’ve processed it, which I find makes it harder to make optimal decisions with those various choices (especially if making critical decisions to reduce noise in shadow areas).
  • Denoise is effectively “free” as it is included with the cost of LR / ACR.
  • Direct integration with LR / ACR.

Pros for Topaz:

  • You can use it on TIF, JPG, and PNG – which is beneficial for improving images you’ve already edited, stock images, etc.
  • You can use it as a filter on a Smart Object. So if you prefer more flexibility, you can change the noise reduction settings later. This may mean an extra Smart Object + filter (if you aren’t applying it to a RAW Smart Object). I’d generally be careful with applying noise reduction after sharpening, clarity, etc.
  • It also offers controls to add some sharpening / detail. This can be very useful (it can also be misleading if you compare the initial Topaz results to the unsharpened results from Adobe Denoise). Personally, I find this may produce unwanted artifact at an early state of image processing which may limit the ability to enlarge later. I prefer not to sharpen this way on the RAW file.

 

I expect Adobe Denoise will spur more innovation and I look forward to seeing how things continue to improve across the ecosystem. If you feel I’m overlooking anything here (or things change in the months to come), please comment below. I’m sure Topaz will keep producing great updates.

 

Other notable changes in LR:

  • Edit in / “Open as Smart Object layers in Photoshop”.
  • Curves in local adjustments.
  • Ability to import  AVIF (new format which is much smaller than JPG) and HEIF (iPhone).

 

ACR v15.3 also has some nice HDR enhancements, including:

  • Vastly improved color for orange/yellow HDR highlights, which is particularly beneficial for images such as sunrise / sunset.
  • Full support for color grading in HDR.
  • A new keyboard shortcut (<shift>-O) to toggle the “Visualize HDR Ranges” overlay.

 

[Disclosure:  This post contains affiliate links.  I have purchased all the software referenced above and only endorse tools I personally use and recommend. If you purchase through these links, a small percentage of the sale will be used to help fund the content on this site, but the price you pay remains the same.  Please see my ethics statement if you have any questions.]

Greg Benz Photography