ON1 DeNoise AI "review"

Will ON1’s NoNoise AI Steal the Crown from Topaz DeNoise AI? Quite Possibly.
(Will On1's NoNoise AI Steal the Crown from Topaz DeNoise AI? Quite Possibly. | Fstoppers)

is a comparison review cited by ON1 in an email advert. I have read the review, and the results in terms of the ON1 application do not correspond to my experience – the application I tried (under purchased license, not “trial”) seemed far more “buggy” than that described in the article. I also note that DeepPRIME did not seem to have been comparison evaluated.

The review reads like an advert: (Excerpt) Where On1 NoNoise has the edge over Topaz is its ability to duplicate layers within the program, which includes blending modes, and then mask the layers using its industry-leading “Perfect Brush." (End excerpt) “Industry-leading” is a very advert statement, and certainly one with which many might take exception. The fact that the ON1 application is better than what Adobe native offers, etc, is no great surprise. As for the fact that DeepPRIME, etc, are AI based only means one is comparing the detailed AI engine (and implementation, as all of these are AI software applications on a non-quantum non-neurosynaptic “classical” computer) and the training set upon which the AI “learned”. If one happens to have an image that is “close” to one in the training set, all the better. At the moment, DxO due to the DxO evaluation of lenses, etc, possibly has the most extensive training set. A query for the DxO marketing group: is DxO using similar outlets to the “review” pushed by a competitor?

Just read the marketing email from ON1 myself.

I use ON1 for the portrait module and for dngs taken on my mobile. Not bothered about NoNoise. I will wait till it is part of the full suite when I renew/upgrade ON1 2021. Topaz suffices for the few dngs I do

I have a license for Topaz Denoise AI 3 primarily for use on non-raw images. When comparing its raw capabilities against DeepPRIME, in my opinion DeepPRIME was superior 100% of the time. On some images Topaz was closer to DeepPRIME in quality than on others but it was never anywhere as good.

I also tested ON1 No Noise. ON1 states that while No Noise can be used for non raw images the quality of the output is still essentially in beta and should improve over time. When comparing Topaz Denoise to On1 NoNoise on raw files I found the output from ON1 was superior and more pleasing compared to Topaz virtually all the time. No Noise also crashed on 3 occasions.

When comparing ON1 to DeepPRIME, l found DeepPRIME to be superior to ON1 NoNoise in every case although the differences were noticeably smaller than between DeepPRIME and Topaz Denoise. No Noise’s main defects were smearing and a greater sense of a plastic look when over applied. This first version of On1 No Noise is a bit buggy and not fully refined but it shows a lot of potential which may be attained in future upgrades.



I have been a Topaz DeNoise user for some time (and more recently a beta tester). I use On1 PR as my main editing software, but tried out both NoNoise and DxO Pure Raw as trials.
I have a lot of images shot hand-held in low light (churches etc.). DeNoise has never been able to deal with this very well, and colour noise turns in to “splotches”. NoNoise fared slightly better in my tests, retaining more detail in black fabrics for example, but still with some slight colour splotching. Pure Raw was better than the other two, with good detail and little or no colour splotching. However, since it only works with raw files, it cannot handle images that are under-exposed.
I used NoNoise in standalone, both single and batch, and by sending images from Photo Raw. I didn’t experience any crashes. I won’t be buying it, but will wait for its incorporation into Photo Raw 2022.

1 Like

Assuming that your camera system provides a true raw (not de-mosaiced) file that PL4Elite (DeepPRIME) supports, I have had no issue with under-exposed images, being capable of doing much more than I could with wet chemistry imaging (film and paper push, pull, dodge, burn). If the image is quite under (or over) exposed (using the channel histograms as the estimator), one often must invest much work making manual adjustments, and in most cases, the yield of “keepers” is not worth (in possible client compensation) the effort. However, since it only works with raw files, it cannot handle images that are under-exposed. However, since it only works with raw files, it cannot handle images that are under-exposed. Note that DNG files come in two forms, true “raw” and de-mosaiced (“de-Bayered”). Only the true raw are accepted by the current Prime and DeepPRIME – at some point I hope that DxO will offer a mode of DeepPRIME that works with de-mosaiced image files. I suspect that this would require extensive “retraining” of the AI that DxO has implemented, and thus may not be feasible from a short-term business case.

I find all their marketing around NoNoise to be like this. It’s simply not a true statement. They also claim ON1 itself is the best RAW processor but again, nope. I’ve said on here before I liked the general UI/UX of ON1 but their processing of my RAW files leaves a lot to be desired. I took a typical image of mine as a test image and gave it the default DeepPRIME treatment. Even though it was not a noisy image, I find a modest amount of DeepPRIME removes the pixel-peepers-only noise that is in fact present and in combination with the lens profiles yields a sharper result — something I am almost always striving for.

I then spent about 15 minutes with ON1 using whatever tools I could find, trying to match the PL4 output and could not get anywhere close with unsharp masks and clarity and whatever other stuff they had. The difference was like viewing an image at half the resolution. Meanwhile, back in PL4, I could push the lens sharpening even further, then add micro contrast boost, and as necessary turn up the DeepPRIME to yield a stunning image.

My comparisons between NoNoise and PL4 on properly noisy photos show a huge difference in favour of PL4. I still marvel at what PL4 can achieve.