Low ISO vs. Denoise + High ISO - Watch This Before You Next Shoot!

If you would like to post, you'll need to register. Note that if you have a BCG store account, you'll need a new, separate account here (we keep the two sites separate for security purposes).

I set my upper to 25,600 on my Z8. It’s quite noisy but without having fast glass it’s a good compromise for me. I will use Photo AI in some of those situations. Sometimes I don’t bother and use other techniques in Capture One. It all depends on the image. Sometimes I’ll have an image where I can mask the background and do some noise reduction and lower the luminance curve. That can sometimes fix a lot. I’m also not super picky. I’m do this entirely fit my amusement and I’m willing to make various compromises when needed.
 
I did once shoot turkeys bedding down for the night in Bosque with a z8 at over 20k iso. They came out ok but you wouldn’t want to print them I would imagine. I find topaz photo ai to be inferior to DxO in high noise photos the results are very uneven. Some ares look ok while other areas are full of artifacts. DxO is more even in my opinion.
 
Thanks for putting together this direct comparison of all the major denoise packages. Can you elaborate on how you setup the options for DxO? I assume you have it set up with the 'force details' slider set to zero and the lens softness compensation turned off? In my experience, the softness compensation adds a lot of sharpening, and if I use it at all I keep it set to 'soft'.
 
It depends the sharpness slider is at zero in the middle. I will often set the Denoise option fairly low, lower in noisy photos if I feel I’m losing too much detail. Then I will push the sharpness til I feel it’s as far as I can go to taste. I’d rather have a bit of sharpening issues than no detail as I can dial it back a bit in LR but it’s harder to pull it back after Denoise has squashed any detail. In these cases I will select and mask in LR or PS and add further noise reduction to the background . I also use the blur tool often as this removes noise as well. I use this method on noisy photos.
 
Many thanks for this Video Steve
You asked what we find…
I shoot Wildlife on Nikon gear using a D6 and Z9.
I am based in Glasgow where at this time of year you are lucky to have sufficient light by 09.00 as its frequently Cloudy and Misty.
My subjects are generally birds, and even on my 600mm f4 FL, images need cropping.
I always use a Gitzo Systematic Series 5 tripod, and a Wimberley Gimbal head.

My strategy is as follows:
I arrive at the hide in the dark.
I start with my D6 until the light improves to ISO 4000, and then switch to the Z9.
The D6 seems better at handling higher ISO than the Z9.
Normally I limit the Z9 to a max of ISO 4000, but am happy to take the D6 to ISO 10000+

I process images in Lightroom and Topaz De Noise ( I am not keen on Topaz Sharpen

PS- I must experiment more with lower shutter speeds.
 
I did it in 4K, but YT is really compressing it. I've had another person tell me that the compression was making it tough to tell.
I am watching it on my BenQ321 on YT in Theater Mode and I changed the default YT (720p) quality setting to 4K. Much easier to see your differences that way. If people haven't done that they may not see the subtle differences.
 
After 3 espresso's and watching the video, I echo the comments of @Anjin San and others above, and mentioned early by Steve, that for many a useable shot is better than a blurry shot. I though have also come to learn from participating here, sometimes I just put the camera down and enjoy the moment.

I have my Z9 set at 10,000 max. Haven't sorted out my Z6iii preferences yet. I prefer LRc Denoise as it means I stay within one ecosystem. Depending on the use and area, I prefer to try Adobe generative fill adding in detail after denoise, over Topaz. Rarely needed for my use cases though. I have not used DXO.
 
Thanks for putting together this direct comparison of all the major denoise packages. Can you elaborate on how you setup the options for DxO? I assume you have it set up with the 'force details' slider set to zero and the lens softness compensation turned off? In my experience, the softness compensation adds a lot of sharpening, and if I use it at all I keep it set to 'soft'.
DxO was at the defaults, the way it would be used as a "one click" sort of Lightroom plug-in. I realize there are a few options in Pure Raw and when I played with them they defaults looked pretty good for this shot. That's the real trick here too - every single shot in this series can be second-guessed. I could have used more texture or sharping with Lightroom, added more or less sharpening with Topaz, used more luminance reduction, less, etc. Any of those criticisms can be lodged for any image I presented with any NR application, straight across the board. However, none of that would have altered the primary thesis of the video - which is when noise overwhelms detail, you can't recover that original information, so low ISO is still better than high ISO with NR (even good NR :) ).
 
In this video, I challenge the statement, "I don't need to worry about high ISO, I have denoise software!"

To what extent is that true? When do you start losing detail? When does AI software start making up detail? Can denoise software really fix everything? Checkout the these tests using Lightroom, DxO, and Topaz for the answers!

We'll also talk about getting better results with high ISO, my favorite guideline for ISO, my software workflow order, and field techniques to improve your chances with high ISO.

Grab a cup of coffee and settle in - there's a TON of critical info in this one - don't take another photo until you see it!

I can only say that it’s an excellent video for anyone seeking information on how to denoise, as well as on the use and comparison of denoising tools. Two thumbs up!
 
I am watching it on my BenQ321 on YT in Theater Mode and I changed the default YT (720p) quality setting to 4K. Much easier to see your differences that way. If people haven't done that they may not see the subtle differences.
That's probably what it was - now that I think about it, YT doesn't always (or usually I suppose) do 4K automatically.
 
My auto on my tablet on YT comes up at 480p, so far from 4k. Setting to 4k just gives the circling the drain icon.
 
I did it in 4K, but YT is really compressing it. I've had another person tell me that the compression was making it tough to tell.
I will let you know if it makes a difference at full screen on my Studio Display…but a, guessing YT is the problem.

(Added later). When I went full screen and selected 4K on my Studio Display the differences are visible. I use Vinegar on my iPad and laptop to help eliminate the mandatory viewing of the ads YT puts in there and haven't figured out how to force 4K there although I was using Theater mode which makes it full window which is essentially full screen on the iPad. On the larger monitor…yeah, the difference is noticeable but there weren't enough non zoomed shots at the various ISOs to really decide whether the differences were there when viewed at Fit to Window…and some of whatever differences there are at 2:1 will disappear when the shot is downsampled for print and even more so for screen.

When you take the shots from the various NR apps and then downsample them to say 1200 for viewing here or whatever size would be needed for say a 16x20 print and then look at those at output viewing size compared to the ISO 500 shot…do the differences really show up there for you? Not recommending another complete video…just a thought on whether the bounds of acceptability for a super picky person change much when you look at output sized resolutions at output size as opposed to 2:1.
 
Last edited:
Really great video Steve. I'm currently assessing my workflow and potentially incorporating LC as part of it. I currently use Photomechanic+ as my ingestion/catalogue solution in conjunction with Photoshop and Topaz Denose/AI sharpener, never been overly happy with Denise / sharpening solution in the workflow so your video was a bit of an eye-opener.
 
An excellent video as always. I think this video is more than just a comparison of noise reduction capabilities and software comparison, it is also a good demonstration of how we should test our own equipment and processing to know our limits. I take from this video that shooting above ISO 2000 with my Z9 I am going to have to be prepared for some loss of image quality. That is of course assuming I do everything else right. The reality is that without adequate light the sensor cannot capture the fine detail. I can remove the noise but there is no detail under the noise. I will have to decide how much this loss of detail, or artificially created detail impacts my image.

I shoot a Z9 as well as a Z6iii, so I am thinking I need to do a little testing with my Z6iii. Does the lower pixel density of the Z6 series allow maybe a stop or more of denoising capability because the image is going to contain less fine detail to begin with or is the opposite true?

I shoot a lot of sports so I have to accept that I am going to be at high ISOs often. What I have learned is that I do need to have multiple apps to remove noise. It was several years ago, and the products have improved significantly since then, but I had a situation shooting sports where I could not use a specific denoise program for one of my granddaughters teams. The fine textures of the material in their game jerseys was interpreted as noise and completely removed leaving very bizarre looking uniforms, another software application handled the images very well.
 
Good video on the effects of high ISO. I agree that sharpness is critical but I believe that both resolution and contrast of the lens and sensor have a big impact on what the software can accomplish. I use either a Z8 or Z9 body with the 400 TC 2.8 lens. I also often use a Tamron 150-500, mostly selected for size and weight. Images from the 400 TC seem respond better to the denoise software, even at similar ISO values. I do agree that each image responds differently. One generalization I see is Topaz does a better job with skin textures like monkey faces or people images. Denoise Pure Raw tends to do a little better with fur and feathers. Lightroom Denoise is just as god most of the time but I use it less because it is quite slow on my machine. I have got good prints up to about 20 X 30 inches at ISO 12,800 using the 400 TC but I think of 10,000 as the top I would use in most cases. I do get get many usable images at 8000 ISO using the Tamron lens. I think of 6400 as the top for that lens. Given the more closed aperture, that limits me to very low shutter speeds as the light drops off using the Tamron. Like Steve says, there are slways trade offs.
 
Last edited:
Thanks for another great video, Steve!!!

You mentioned in the video that Topaz PhotoAI tend to over sharpen our images, yes, I fully agree. Moreover, some previous versions are better than the new ones...
I am a Windows 11 PC user, and I found a workaround to get better sharpen results (recovering detail from high ISO photos) with PhotoAI. Here it is my workflow:

1- Do minimal light adjustments in Lr (no contrast, dehaze, clarity. etc)
2- Bring the Lr sharpen slider to zero (Detail tab)
3- Send the image into Ps (Edit in Photoshop)
4 -In Ps, I send the photo (layer) to Topaz PhotoAI and usually accepet the adjustment levels proposed by Photo AI (Denoise and Sharpen)
5- I send the photo back to Ps as a new layer and continue my editing from here (in Ps and/or Lr).

With this workflow I usually get acceptable level of sharpening.

The PhotoAI sharpening preview shows way too much sharpening that it is not true when you send the image back into Ps. I think that the PhotoAI image preview is low quality in order to get faster processing...
Even though, sometimes I do denoise and sharpening separated in PhotoAI (denoise first and sharpening after). This way I get more control in Ps of the sharpening level produced by PhotoAI by reducing the opacity slider of the sharpen PhotoAI layer.

With these methods I get acceptable/good results by 90% of my high ISO images. It goes without saying that for the ISO lower end in the high ISO category, I prefer to use the Steve method in Lr.
 
Last edited:
Great video Steve.
I've gone from being a heavy Topaz user to a mix of Topaz/PR and now have gone to a 100% LR AI user. I'm lazy and just find it easier to stay in LR all the time. Every once in a while I'll download the latest trial version like I did for the recent PR and compare to LR and I just don't find any significant differences to make it worth it to me to pay for these other programs anymore.

I'm curious what others have found with DXO PR....I found that I had to turn all the optical correction stuff off or else the image was way over sharpened. Are others finding they can use the lens correction settings and still have a good looking image?
 
Great video Steve.
I've gone from being a heavy Topaz user to a mix of Topaz/PR and now have gone to a 100% LR AI user. I'm lazy and just find it easier to stay in LR all the time. Every once in a while I'll download the latest trial version like I did for the recent PR and compare to LR and I just don't find any significant differences to make it worth it to me to pay for these other programs anymore.

I'm curious what others have found with DXO PR....I found that I had to turn all the optical correction stuff off or else the image was way over sharpened. Are others finding they can use the lens correction settings and still have a good looking image?
Yep. I set the sharpness correction to soft, but otherwise generally speaking it's a good starting point at the defaults. Some images require changes, but not often.
 
I think the defaults of Photo AI for sharpening are a bit extreme at times but I just spend a minute or two tweaking the strength and model I use and it usually behaves pretty well for me. I think the learned that a lot of people want the slightly over sharpened look just like how people liked the extreme HDR look years ago. The key is to work it. You can also change it by specifying areas more carefully instead of letting it pick. My guess is that all of these tools could produce a very similar result with some time tweaking the parameters.
 
Thanks for another great video, Steve!!!

You mentioned in the video that Topaz PhotoAI tend to over sharpen our images, yes, I fully agree. Moreover, some previous versions are better than the new ones...
I am a Windows 11 PC user, and I found a workaround to get better sharpen results (recovering detail from high ISO photos) with PhotoAI. Here it is my workflow:

1- Do minimal light adjustments in Lr (no contrast, dehaze, clarity. etc)
2- Bring the Lr sharpen slider to zero (Detail tab)
3- Send the image into Ps (Edit in Photoshop)
4 -In Ps, I send the photo (layer) to Topaz PhotoAI and usually accepet the adjustment levels proposed by Photo AI (Denoise and Sharpen)
5- I send the photo back to Ps as a new layer and continue my editing from here (in Ps and/or Lr).

With this workflow I usually get acceptable level of sharpening.

The PhotoAI sharpening preview shows way too much sharpening that it is not true when you send the image back into Ps. I think that the PhotoAI image preview is low quality in order to get faster processing...
Even though, sometimes I do denoise and sharpening separated in PhotoAI (denoise first and sharpening after). This way I get more control in Ps of the sharpening level produced by PhotoAI by reducing the opacity slider of the sharpen PhotoAI layer.

With these methods I get acceptable/good results by 90% of my high ISO images. It goes without saying that for the ISO lower end in the high ISO category, I prefer to use the Steve method in Lr.

A plus to that is you can adjust the opacity of that layer as a fine tuning. I do similar but go from lightroom to photolab then to photoshop and use ctrl-alt-e to stamp all visible, convert to smart object and apply the Topaz sharpen filter as output sharpening.
 
Steve, Thanks for the video and all you do. I resonate more with the field tips a LOT.

I have intentionally avoided all three of the tools in this video. It is not that any of them are bad/good/better/worse.... Maybe it is the contrarian in me... not sure. There has always been something about the "advanced NR software marketing" that rubs the cat the wrong way. When rubbed the wrong way, I try to understand the root cause of the issue to come up with alternate (hopefully better, but not always) methods. [Certainly, a bit of a personal psychosis]

For me, the favorite tip from your video is filling the frame. Mathematically, a big deal. It always makes me cringe seeing Noise Reduction methods try to correct for extreme cropping.



I will call the following... "Getting the best out of NR software"

Here are some adjacent thoughts that I have found helpful. YMMV. But I think they apply across all Noise Reduction tools.


Filling the frame also helps using matrix for autoexposure.
For example, do you really want the bird's eye setting the exposure or do you want the feathers to set the exposure? Personally, I like the bird eye for focus, and the subject (bird?) body for exposure. Filling the frame helps a matrix auto-exposure do its work. Better is probably manually setting exposure (knowing when to push exposure), but I am sometimes too lazy or too busy to this in the field.

Glass
Somewhat adjacent to your recommendation on starting with the best possible focus.
Softer glass hurts the detail... in turn, less detail to the Noise Reduction software to use. My worry about noise has decreased as I have afforded improvements in glass. To help me understand root cause, I experimented with the multiple ways to achieve the same 135mm focal length. [135mm selected is because that is what I have on the shelf] At the low end, an adapted 1970's 135mm/2.8 Zuiko prime; in the middle, a couple of zoom lenses; at the high end, a 135mm Plena. For the same ISO and f/stop, the difference in higher ISO noise post processing can be night and day. Would like to do the same experiment with the 400mm/2.8, but I need to sell a kidney first. LOL.

Denoise with masks...
To me, the delta in noise recovery between different areas in an image is almost as important as the noise recovery itself.
Large out of focus areas tolerate (appreciate?) stronger noise reductions (machete). While the eyes and feathers or other details seem to appreciate the scalpel. Thus, applying the right NR at the right location seems to help the final result.

Help NR software by using the most bits possible.
I do tend to stay away from noise reduction processing 8-bit images. Thus, tend to use RAW/TIFF to get greater info into the NR tool.

Mathematic image stacking to base level random noise to better information for later Noise Reduction processing.
If the subject doesn't move within three or more images taken at a 20fps frame rate, overlaying multiple shots with the right mathematics takes advantage of the random noise. Thus, providing a better starting point for noise reduction software (sometimes, removing the need for noise reduction software entirely)
 
In this video, I challenge the statement, "I don't need to worry about high ISO, I have denoise software!"

To what extent is that true? When do you start losing detail? When does AI software start making up detail? Can denoise software really fix everything? Checkout the these tests using Lightroom, DxO, and Topaz for the answers!

We'll also talk about getting better results with high ISO, my favorite guideline for ISO, my software workflow order, and field techniques to improve your chances with high ISO.

Grab a cup of coffee and settle in - there's a TON of critical info in this one - don't take another photo until you see it!

Great video, Steve!
 
Excellent overview and help to understanding noise reduction VS ISO.

I don't use Lightroom (like it seems , most people do...)-have always used Photoshop, Recently Tim Gray presented a tutorial on Adobe Camera Raw, which I have found very useful . It is a far cry from what it used to be and includes sharpening and noise reduction.
In my processed version, I used Camera raw for everything. Then in PS I selected the background and used a little gaussian blur . Curious what you all think of this process.
I must say, Adobe Camera raw seem very sophisticated and I've very touched the surface of what it can do. But I like it better than DXO or Topaz for the most part.

 

Attachments

  • milkweed.orig..vs.DSC_0228-2.jpg
    milkweed.orig..vs.DSC_0228-2.jpg
    65.7 KB · Views: 19
  • milkweed.processed.vs.DSC_0228.jpg
    milkweed.processed.vs.DSC_0228.jpg
    71.1 KB · Views: 20
Denoise with masks...
To me, the delta in noise recovery between different areas in an image is almost as important as the noise recovery itself.
Large out of focus areas tolerate (appreciate?) stronger noise reductions (machete). While the eyes and feathers or other details seem to appreciate the scalpel. Thus, applying the right NR at the right location seems to help the final result.

That's the basis of my Lightroom Denoise video :)

Mathematic image stacking to base level random noise to better information for later Noise Reduction processing.
If the subject doesn't move within three or more images taken at a 20fps frame rate, overlaying multiple shots with the right mathematics takes advantage of the random noise. Thus, providing a better starting point for noise reduction software (sometimes, removing the need for noise reduction software entirely)

Agree - I'm thinking of doing a video about it, although I have to brace for all the "How the heck do you get an animal to hold still" comments. I've tested this method before and it does work well - really well in combination with NR. I have found I needed more than just a few frames though - I think I had to have like 5 or 6 before it really became impressive.
 
An excellent video, Steve. Appreciate all the time you dedicate educating people to improve their photography skills! It occurred to me that when shooting at high iso we don’t have the benefit of a low ISO image for comparison. At high ISO detail is lost and can’t be retrieved. So it would be interesting to process an image using these three programs and picking one you prefer, without a standard for comparison. Introducing a low ISO image can introduce observer bias. Picking the image that looks best may or may not correlate to the gold standard taken at low ISO, but may indicate how well the processing attempts to compensate for lost details.
 
Back
Top