ON1, a longtime images software program firm situated in Oregon, just lately launched its new noise discount utility known as NoNoise AI. As a longtime person of Topaz Labs DeNoise AI, it solely made sense to pit these two machine-learning-based noise discount packages in opposition to one another and see which one comes out on high.
Earlier than diving right into a comparability of ON1 NoNoise AI and Topaz Labs DeNoise AI, there are a few issues to level out. First is an inventory of seen destructive results from photographing with excessive ISOs in order that we could extra intently study the outcomes beneath.
- Much less element and fewer sharpness
- Diminished dynamic vary
- Colour shifting and colour fading
- Scorching pixels
- Blotchiness or smearing, normally from in-camera noise discount
Whereas these two functions primarily goal “digital grain” kind noise, there are different degradations occurring to a picture concurrently because the noise will get worse and, to an extent, I seen each software program functions trying to repair just a few of those as nicely, so I assumed it was price mentioning.
One other factor to level out is what the “AI” in NoNoise AI or DeNoise AI is doing. The packages should not merely discovering hidden particulars already current “behind” the noise and enhancing them. Simplified, the software program is studying your picture, evaluating it to its large knowledge library complied from machine studying to acknowledge what one thing ought to appear like, after which replaces the unique artifacts with clear textures and particulars primarily based on its photo-driven libraries.
In observe, you don’t actually take into consideration this and to the end-user, it does simply appear like unique particulars being cleaned up and introduced out as a result of it really works so nicely (more often than not). Nonetheless, organizations just like the Nationwide Audubon Society, which simply hosted the 2021 Pictures Awards Contest, acknowledge the distinction and particularly didn’t permit “AI and machine-learning-based software program used for upscaling, sharpening and important noise discount.” In a way, it may be argued that the ultimate edited pictures from AI noise discount software program are composites.
Options and Performance
Shifting into the comparability, I need to first talk about the similarities and variations so far as the structure, instruments, and file dealing with workflow goes for every utility.
Opening every program, each have an identical structure in default settings. The picture being processed is oriented on the lefthand facet, the principle toolbar panel is on the precise, and there are extra info and instruments on the highest and backside.
So far as seems go, ON1 NoNoise AI is by far superior with its minimal use of colour. Topaz Denoise AI however tries its finest to skew your colour notion by sustaining a blue colour palette all through this system — a giant no-no for the modifying course of.
Each permit colour modifications to the preview background colour however, for probably the most half, that is by no means seen except your zoom is backed out all the way in which and it additionally doesn’t have an effect on the panels surrounding the picture. ON1 goes a step additional to supply a customizable accent colour, however this makes a minimal distinction to something.
Trying on the primary software panels, we see a common similarity in that there are instruments for noise discount and instruments for sharpening.
Topaz Labs has a further AI Mannequin part that has 4 choices: Customary, Clear, Low Gentle, Extreme Noise. Customary can be in your common images with low to reasonable noise. Clear makes use of an older Topaz Labs noise discount engine and doesn’t depend on AI as a lot, and this mode is sweet for occasions whenever you see a whole lot of inconsistent noise discount and artifacting launched by the AI fashions. Low Gentle and Extreme Noise type of sound like they’d be focusing on the identical factor, nevertheless, in case your closing picture is supposed to have a whole lot of darkish areas it is perhaps finest with Low Gentle.
Whereas Topaz Labs has one slider for noise discount and one slider for sharpness on the coronary heart of its program, ON1 has three sliders for each and permits for extra refinement, which is vastly appreciated. It’s extra hands-on, but in addition extra time-consuming. Each have Auto toggles that are useful to get a suggestion on the place to begin with the sliders. From my lengthy expertise with Topaz Labs, I do know that typically the Auto suggestion is totally off base, however it normally will pace up the method.
One of many extra attention-grabbing sliders that exists in ON1 NoNoise AI is the colour software (and to make certain, no, this doesn’t act just like the Colour Noise Discount software in Topaz Labs). I discussed earlier in my listing of destructive results of excessive ISO taking pictures that color-shifting turns into an issue. Colour noise may give off a shift within the blacks and different colours get all pale and skewed. This good software works to handle that by offsetting these points. Generally in a decrease ISO picture, the software isn’t doing any favors when absolutely energetic, however for actually noisy pictures it makes an enormous distinction.
For instance, beneath is a photograph taken with the Sony Alpha 1 at ISO 25,600.
The urged workflow for every program differs in that ON1 NoNoise AI prefers to work off of unique RAW information whereas Topaz Labs DeNoise AI is generally set as much as take processed information. Topaz Labs simply launched model 3.2 which supposedly consists of enhancements to the RAW colour processing, nevertheless from what I may see evaluating v3.1 and v3.2 was a lens profile correction and that’s all. As for ON1, whenever you attempt to import a non-RAW file, there’s a pop-up message that claims, “outcomes from non-RAW images are thought-about beta high quality and can enhance quickly.”
Topaz Labs simply wins right here, as it’s extra versatile and leverages packages like Seize One to do a much better job on the RAW processing. ON1’s logic is that noise will get launched throughout the demosaicing stage of RAW processing and due to this fact they’ll get to it on the supply. What’s misplaced, although, is the unimaginable colour processing and RAW-level changes from packages like Seize One which aren’t the identical in any respect after ON1 outputs a transformed DNG.
One different factor to say because it pertains to workflow is that each of those functions have superior masking talents to regulate the place the noise discount is being utilized, nevertheless, solely ON1 may also do layering inside the program. This is usually a large benefit to those that want various localized noise discount and sharpening with out having to take care of going out and in of Adobe Photoshop a number of occasions.
One different factor to have a look at is how previews are dealt with. Totally processing the images on the finish of the method can take a while, so previews permit customers to get an excellent however not good instance of what the picture will appear like. I discover that ON1 previews are snappy and replace in close to real-time when dragging the sliders. The draw back is that the preview accuracy in comparison with the outputted picture is just what I’d name “first rate.” However, Topaz Labs previews take longer to course of and replace, however the accuracy is near the identical because the outputted picture.
Since I’m the one doing the processing, right here’s my idealistic outcome to remember: I consider noise discount is a balancing act. Utterly wiping the noise away till the picture is a smeary mess of zero element is simple — any modifying program can try this. What I aimed for in my testing was attending to the purpose the place the noise was lowered as a lot as attainable whereas additionally retaining or enhancing as a lot element as attainable within the picture with out getting “crunchy.”
This typically means I like a bit of little bit of noise left within the picture so it doesn’t look plasticky and doesn’t wipe away the high-quality particulars that profit from a bit of grain-type noise. Noise current within the outcomes beneath isn’t a sign that the packages can’t obliterate it, it’s a mirrored image of me attempting to get the best-looking outcomes whereas toeing the road of fresh, but detailed.
Within the instance pictures beneath, I’m utilizing ON1 NoNoise AI model 188.8.131.5285 and Topaz Labs DeNoise AI model 3.2.0.
What I discovered right here was that I used to be in a position to get extra high-quality element out of Topaz Labs with this comparability. Nonetheless, ON1’s Colour slider was in a position to eliminate the excessive ISO colour forged occurring within the picture.
On this comparability, you possibly can see that neither ON1 nor Topaz Labs have any good purpose to be touching the RAW information. Each are trying sickly when modifying off the RAW in comparison with Seize One. I say, let the RAW processing be dealt with by the professionals.
Whereas I believe it’s simple for the eyes to go for Topaz Labs right here as a result of it’s extra vibrant, in the case of noise discount I believe ON1 was in a position to clear extra colour noise within the darkest areas. It’s only a disgrace about its RAW colour processing.
This one left me speechless at how superb ON1 NoNoise AI might be. Take a look at all that high-quality element introduced into the deer’s fur whereas nonetheless eradicating the noise. Topaz Labs ended up getting muddy particulars that I wasn’t in a position to finesse into something higher. There are components the place NoNoise AI will get oversharpened for my part, however with masking and layers inside the applying, it may be toned down on an area foundation.
As soon as once more, ON1 is ready to clear the colour noise higher than Topaz Labs. It additionally brings out particulars higher, however it does teeter on being too sharp in some areas. It’s one other spectacular efficiency from the brand new ON1 NoNoise AI when Topaz Labs DeNoise AI has been round for a few years.
It’s Within the Particulars
In the present day, my winner is Topaz Labs DeNoise AI for being extra versatile in the place I can use it in my workflow whereas getting good outcomes a lot of the time. I discover the RAW colour processing facet of ON1 NoNoise AI to be poor, and it will get worse when Seize One doesn’t play nicely with the transformed DNGs as soon as imported again in in comparison with the digicam producer’s proprietary RAW file.
That mentioned, apart from some quirks of utilizing ON1 NoNoise AI — like its insistence on at all times defaulting to the annoying split-screen slider and leaping zoom ranges after I didn’t ask it to — I want its toolset and structure extra. I really feel like I’m actually in a position to hone in on what I have to do to get the most effective outcomes. I believe that after ON1 learns to take care of already-processed RAWs from a superior processor (it’s unusably unhealthy whenever you strive it proper now) then it is going to take over as my go-to noise discount software program.