In this post I want to share some experience with photo processing using ON1 NoNoise / Photo Raw 2024 (still the current version of the AI tools). There are now many programs out there with AI de-noising, sharpening and more capabilities and they likely all do a good job. ON1 allows quite flexible tuning of their algorithm and for my images the default settings are NOT what gives me best results. This may be comparable in other programs – so here it comes.

My personal “sweet spot” is to use both, NoNoise and TackSharp. For NoNoise, I set luminance to 65%, detail to 23% and color to 50%. Tack sharp is very aggressive and I only add it in to give a small “touch-up” at 15%. There is another slider called “microsharpen” and I tend to leave that at the default of 100. Like this:

If the contrast is too crisp, I reduce the microsharpen value (to 50 or even 30). If I am disturbed by AI artifacts I first reduce the TackSharp to 10% and then if necessary the detail to 15 or 20%. Usually that is enough. I use either an Olympus E-PL5 or an EM-5 (the first model). Both cameras have the same sensor, so the IQ is the same. And they certainly benefit from modern processing software.

Examples, of course!

I start out with above-water pictures because I vary the ISO more than under water. But water is close (lens was a Lumix 12-32 kit zoom, the tiny one).

First, an outside picture at base ISO of 200 (full image is AI-processed):

And a 100% crop without any de-noising applied (so essentially raw but with adjustments of contrast, vibrance etc.)

Here is the crop with the AI-settings applied:

I like the detail in the shadow below the deck that is much better here; some artifacts are visible, but not too bad.

 

And now an image of the steam engine inside the boat, this was taken at ISO 6400 (full image is AI-processed):

Here’s a 100% crop with no noise reduction applied. Not bad at all, I think – considering it is a 13 year-old camera model.

I use ON1 PhotoRaw and this also offers a “classic” de-noising option so I can compare NR options while everything else is the same. Probably comparable to other programs and I applied this with caution at something like 25% luminance, 50% detail, 100% color and 50% detail in that.

Slightly better, but if you go stronger luminance values it wrecks all the detail.

Now here is the crop with the AI settings applied:

Not magic – but better still. There is a bit of luminance noise left, but for my purposes this result is perfectly sufficient.

What I like about this algorithm is the “elasticity” – both images were AI-processed with the exact same settings and they both (and many more) turn out nice. The program defaults are way too aggressive, though. As mentioned before these settings are my “sweet-spot” and I now routinely apply this to essentially all images (but read the very end of the post).

Does this turn my classic cameras into a current Sony full-frame model with a Zeiss Batis lens?

Certainly not. But this is better than the out-of-camera jpg (which I am not showing because contrast ect. is different). And it of course does not change the autofocus-performance of my old cameras - just forget about using TackSharp to recover an image that is clearly out-of-focus. That’s a myth.

Now what about under water?

This is a grouper in Madeira, shot with the E-PL5 and the 14-42 kit lens behind a flat port and an Inon UWL-H100 wetlens (no dome) at ISO 200; some sand was in the water and I used strobes, so lots of backscatter (full image is AI-processed):

A 100% crop of the cheek with no noise reduction applied:

Here is the “classic” NR version (though it really doesn't need any NR):

And here comes the version with AI applied (my settings):

Some detail on the grouper skin is probably AI fantasy, but if I really turn up the contrast on the original I can see the larger wrinkles. The grains of sand were real, too. Artifact or not – once I stop pixel-peeing, it looks really natural and I don’t have the actual grouper to compare with. But I do think that ON1 was trained in part on pics of leather sofas.

This is the eye with no NR:

With classic NR:

And with the AI settings:

Pick your preference…

I am also cropping a region with backscatter for obvious reasons. For this, I am comparing the mild “classic” noise reduction:

And the AI version:

I read they trained the model to be well-suited for starry skies, and this can be seen. Backscatter is more focused and more intense – which is, after all, what I am asking for with my settings. In this picture, the backscatter has to be removed by other means anyways; I ended up using a blur and masking out the grouper. Whether it is easier to do this before or after applying the AI de-noising I honestly cannot tell, it probably depends on the strategy and tools employed.

Is there any other downside to this?

Yes, and that is processing time. I do not have a powerful computer (a 5-year-old laptop with an i5 processor and its built-in graphics only). Processing a 16 MPx raw image with these settings takes literally 15 minutes. To make things worse, ON1 PhotoRaw will go through a new calculation every time you re-open the image for processing (I hope the Photoshop-plugin is different). Thus, with the settings in place, it takes me 15 minutes to open an image. No kidding.

That’s why I don’t apply these settings until the very end. Rather, I open images without any NR, adjust whatever I need (contrast, white balance, remove blemishes etc.) and with highlights and shadows I leave a tiny amount to room because the AI procedure increases contrast a bit – thus I avoid clipping. When I am all done with the images in e.g. one folder, I navigate to a different source image, copy out the NR sweet-spot settings, go back to the current folder, select all images, paste in only the NR settings, select to export all images as high-quality jpg to a subfolder and then go to bed. Depending on how fruitful the trip was, I might even go to bed twice before it’s done. This is why the “elasticity” of the algorithm is so important & useful.

If you want to try it out – they offer a free 30-day trial with no restrictions, I believe. There are standalone packages and a photoshop plugin. I have no affiliation whatsoever with the company and I am sure other tools can give comparable results. Just make sure you venture beyond the default settings.

Edited May 17May 17 by Klaus
bad click too early

Wow, thanks for the detailed review and tutorial.

Good to know we have all these tools available, and it's our call on how and where to best use them.

Thanks indeed. Amazing technology but, yeah, comes at a price.

I’m using similar tools on LR on a fairly basic Mac Studio with an M1 processor. That is pretty quick and can handle the AI “Denoise” in under minute. The results can be really impressive and have resurrected images that would have been right offs in the past.

The computing issue can be solved by either money or patience, if you don‘t need to be quick on social media. I have not had much luck with recovering images from the bin, but I like using this to „forgive“ slight optical imperfections from kit lenses etc.