Recently we looked at how Topaz DeNoise AI compared to other leading noise reduction software and found it to do the best job of both removing noise and keeping detail in photos. This article will take an in depth look at how to use Denoise AI. If you haven’t seen our comparison yet you can read it at:
Topaz DeNoise AI Compared
Note that this article has been written for v2.1.6. It will be updated if Topaz introduces any changes to future versions.
A Quick Sample
First let’s take a look at a sample image. Below is an ISO8000 file from a Nikon D500 at 100% view. You can see how much cleaner the processed image (right) is than the original (left). I have labelled three areas where you can really see the difference DeNoise has made in both reducing noise and retaining detail.
ISO8000 from a Nikon D500 at 100% zoom. Original image is on the left, DeNoised is on the right.
When you first run Topaz Denoise AI the software will run some initial checks in order to optimise it’s settings. You should double check that these are correct for your system in order to maximise performance. Open the preferences by clicking
File -> Preferences on a PC or Topaz DeNoise AI -> Preferences on a Mac.
The advanced preferences will determine which hardware in your PC is used to process your images.
Enable Discrete GPU: This should be ‘Yes’ if you have a modern Graphics card in your PC or attached in an external GPU setup. The graphics card will be used to process the images instead of the computers CPU and will (depending on the card) be faster.
Allowed graphics memory consumption: Higher settings will allow for faster processing, but may mean you won’t be able use your PC while processing images (everything will slow down to a crawl)
Enable Intel OpenVINO: If you set ‘Enable Discrete GPU’ to ‘No’ and you have a modern Intel processor you should enable this option. It will greatly speed up processing times.
You should also check these if you are getting any unexpected results after running the software. Incorrect settings can cause issues with certain hardware.
Prepare Your Image
Topaz Labs recommend that you run DeNoise as the very first step in post-processing. However we have found this not to produce the best results. As a first step you should apply any tone and colour adjustments in your favourite editor before sending to DeNoise AI. You can read more about our suggested workflow in our article ‘Topaz AI Workflow’
Just be sure to set any noise reduction and sharpening levels to zero at this stage of the editing process. If you’re processing with DeNoise AI it is better to let it handle 100% of the noise reduction. Sharpening can be applied in DeNoise as well, or should be applied after the noise reduction process.
Once that is done you should send the image to Topaz Denoise AI to get started on noise reduction.
Sending an image to DeNoise AI from PhotoShop
If you are comfortable with the DeNoise AI interface you can skip the next section and get straight to the Noise Reduction Settings
The View Settings
The default view settings in DeNoise AI are set up quite well, but you can change how things are laid out with with the view settings described here. I have settled on using the Side-by-side view for my work, but its up to you to decide what you prefer.
Click on the View icon to change the preview window
Side-by-side View (default)
This will show the original, unprocessed image on the left and a preview of the processed photo on the right. This view makes it very easy to compare how the final image will look compared to the original.
Note that the ‘Original’ button is not active in this view as the original image is always shown on screen.
Side-by-side view shows the original image on the left and the processed one on the right
The split view will show only one copy of your image on screen, but has a divider that you can move to ‘uncover’ and compare the original image to the processed one. Anything to the left of the divider will show your original image, to the right is the processed version. You can drag the slider to the left and right to easily compare the processed image vs the original.
Clicking and holding the ‘Original’ button will show only the unprocessed image on screen. As soon as you left go of the mouse button the split view will return.
Split View allows easy comparison between the original image (left of the divider) and the processed on (right of the divider).
When Single View is selected you will only see the processed image. If you want to compare it to the original you can click and hold the ‘original’ button in the toolbar. Holding the button will replace the processed image with the original one. Letting the button go will bring the processed image back.
Single View shows the processed image. You can hold down the ‘Original’ button to view the unprocessed image.
Brightness and Zoom
Clicking the ‘Brightness’ button on the toolbar will increase the brightness of any really dark areas in your image. I don’t find this setting very useful, but it is designed to make any noise in dark areas more obvious so you can better see the effects of noise reduction.
It is important to note that turning the ‘Brightness’ setting on does not affect the final output – it is merely to help reveal the effect of any noise reduction on dark areas of the image.
The zoom option does exactly as you would expect. On small high resolution screens (such as a 4k laptop screen) it can often be helpful to zoom in to 200% in order to better see the noise in your image. Most of the time you will want to leave this at 100%.
Navigator and Auto-Update Preview
The navigator shows a thumbnail of your entire image. The white box inside the image represents the area that you can currently see in the main preview screen. Holding and dragging your mouse inside the box allows you to move the preview to the section that you desire. You should should always check a few sections of your image before saving it. I recommend checking a sample of both detailed and smooth areas to make sure you have chosen the correct settings. The aim is to balance smooth backgrounds with as little loss of detail as possible.
If you don’t have ‘Auto-Update Preview’ selected (discussed below) then you will need to press the ‘update’ button to show the processed image after you move the navigator around.
The navigator window allows you to reposition the preview window. Just drag the white rectangle around.
Checking this option means that every time you make a change to the noise reduction or sharpening sliders, or move the preview area in the navigator, the preview image will process and update. This can be very time consuming when needing to update multiple sliders at once. If you untick this option the preview will only update when you click the ‘update’ button.
I find that the manual update method works best for me as having to wait 10 seconds or so for the preview to update each time I move a slider can be quite annoying. You just need to ensure you press the update button each time you want see how any changes will affect your final image.
Noise Reduction Settings
Now we will go through the important bits: the settings that will remove the noise from your photos
Noise Reduction Models
DeNoise AI can use one of two methods to reduce noise. DeNoise AI is the default mode and will give the best results in most cases. AI Clear is an older model that can be useful if DeNoise AI gives unexpected results. DeNoise AI will generally preserve more detail in the processed photo that AI Clear and is the recommended model to use in most cases.
Ticking the auto-detect settings box will force DeNoise AI to analyse your image and choose noise and sharpening settings (discussed below) that it thinks is best. Topaz normally does an OK job here and I will usually tick the box to get an initial idea of what my settings will be, then adjust from there.
This is the slider that tells Topaz DeNoise how aggressive to be when removing the noise in your image. Higher numbers will remove more noise, but with the potential to also remove some image detail. You should choose the lowest number you can that still produces a noise-free (or close to it) image.
Don’t forget to hit the ‘Update’ button to see the changes if you have ‘Auto-update Preview’ turned off (as discussed above).
Adjusting the Remove Noise slider will change the strength of the noise reduction
It is important to note that the complete removal of noise is not always the best option. A moderate amount of noise will probably not show up in standard print sizes or when viewed on a screen. It is almost always better to have a highly detailed image with a small amount of noise than a noise-free image that has no detail due to strong noise reduction. Remember that your image will rarely be viewed at 100% zoom levels.
This slider tells Topaz DeNoise how much sharpening to apply to your image. Doing initial sharpening at the same time as noise reduction will often have better results than applying it at a later stage.
Once again, don’t forget to hit the ‘Update’ button if you have disabled Auto-Update preview if you want to see the effects of any changes to the sliders.
Adjust the image sharpness using the Sharpen slider
As with the noise reduction slider – the aim here is to be fairly subtle. The resultant image should be sharp, but avoid ugly sharpening artefacts by apply too strong of an effect.
Even with the slider set all the way to zero, DeNoise AI will still apply a base amount of sharpening to your image. There is no way (currently) to apply noise reduction without adding some sharpening to the image.
Recover Original Detail
This slider will bring back some of the detail from your original image. It will re-introduce some of the original noise back into the photo and can help avoid the over processed plasticy look.
Use the Recover Original Detail slider to bring back some grain and get a more natural look.
I usually experiment with this slider in combination with the Remove Noise slider to find the combination that works best. For most images I find a value between 5-10 will have the desired effect.
Color Noise Reduction
This slider deals with colour noise that can be found in high ISO photos. If you notice any blotchy patches of colour in your image after the other noise reduction has been applied you can remove it with this slider. I rarely find I need to add any colour noise reduction in Topaz Denoise and leave it at zero for most images.
Save Your Image
Once you have set the perfect noise reduction values you can now press the ‘save image’ or ‘apply’ button to apply the results to the whole image and save it. Depending on your hardware and the size of the image you are processing this can take between a couple of seconds to several minutes.
You will the ‘Apply’ if you have opened DeNoise from another application. It will say ‘Save Image’ if you opened the standalone editor.
If you sent the photo to Topaz via a host application (Lightroom, Photoshop, On1, etc), DeNoise AI will close itself when it is done and return you to the host application. If you are using the standalone program your processed image will be removed from the interface ready for you to process another image.
Introduced in version 2.0, you can open several images at once, choose the ideal settings for each image (or leave it on ‘auto’ if you desire) and process them all in one go. This can be useful if you have a bunch of images to process that will take a long time. You can set it up, press ‘save’ and do something else while you wait for the images to be processed.
You can open multiple images by dragging them from your PC file system into the DeNoise window, or using the File -> Open Images menu item.
Batch Processing allows you to work on several images at once.
So, now you know how to process your images with Topaz DeNoise AI. Once you get used to the interface it will probably take you about a minute per image to get the settings right. You will mainly only need to adjust three sliders:
Remove Noise: To adjust the strength of noise removal
Sharpen: To apply subtle sharpening to your image
Recover Original Detail: To bring back a small amount of grain for more pleasing images.
The actual processing of each image will take some time. Between 10 seconds and a couple of minutes depending on how large the image is and how fast your CPU/GPU is. The results this software is capable of can be well worth the wait.