DeNoise AI vs AI Clear

I find that AI clear works better on heavy noise while Denoise AI may work better on light to medium noise. Its goo to have both in one package however i use AI Clear in many presets I have made and I don’t want to lose that in Studio.
Here is an example of both algos in Denoise AI. i have circled areas where AI Clear has made things worse. This was a jpeg photo (very little noise) and the Denoise AI stand alone v1.1.1 was used for both.
AI Clear:

Denoise AI:

I tested a photo in both Studio with AI Clear and then in Denoise AI using both the ‘native’ and AI Clear algorithms. Interestingly, the Studio AI Clear result was much superior to the Denoise AI result using the AI Clear algorithm. And as expected (by me anyway), the Denoise AI result using the ‘native’ algorithm was better than the Denoise AI using the AI Clear algorithm. In the end, the results (in my opinion) from best to worst were: Studio + AI Clear, Denoise AI + native, Denoise AI + AI Clear.

I suspect this will be the case for the foreseeable future in so much as our results will vary as new models are introduced. This is the upside/downside to the use of ‘AI’ and models… instead of a specific algorithm that is consistently repeatable with expected results across a variety of photos, the AI models will vary in consistency across the same photos.

Again, I applaud Topaz for embracing the use of learning models and look forward to improvements in these products over time.

1 Like

I am experiencing exactly the same thing. AI Clear in Studio consistently produces better results than AI_Clear in DeNoise_AI.

2 Likes

So there are two possibilities…

The first is that the AI Clear model was intentionally manipulated to produce an inferior result than the ‘native’ model in Denoise AI in order to boost the ‘native’ model’s appeal.

The second is that the AI Clear ‘model’ itself is in fact being utilized however the Denoise AI code (algorithm) that executes it is different than the code that executes it within Studio.

I suspect that the second possibility is more likely. There is a reason that the new AI products are stand-alone - they are completely different code bases (from Studio) and they appear to share much of the same code for the user interface, etc. If you recall the history of Topaz products, they have had stand-alone products in the past (like Remask, Impression, Texture, etc.), integration products like photoFXlab, Studio (that integrated many of the previously stand-alone products) and now back to stand-alone products. There is both a business imperative to do this (sell more products) and the reality that the vision of Topaz is evolving/re-factoring and along with it, its development team and methodologies. This is largely acknowledged in one or more Topaz blogs. We see this with other companies such as Adobe and their evolution of Lightroom.

Just my speculation based on 30+ years of software development myself… still, that does not solve our noise problems! The good news is that Topaz has embraced the use of machine learning and is actually putting it to use instead of using the word ‘AI’ to simply sell products. What that means though, is that it will constantly improve over time as new models are trained. I don’t know if this is happening now, but I would be fine having the Denoise AI product (optionally) send data about the photos I am working on to help continue to improve the models - I noticed when looking at preferences, that Adobe seems to be doing the same thing in their products.

As I hope Topaz reads these posts - please know that we all support your products and strategy and are committed to your products/company in the long term - however, until these products are consistently processing our photos, please continue to provide the ability for us to use existing products (like Studio + AI Clear) that work well for us today.

2 Likes

I also confirm that “Studio+AIClear” is MORE better than “DenoiseAI+AIClear”!

But, also, after the latest updates, “DenoiseAI+DenoiseAI” has become better!

“Studio+AIClear” and “DenoiseAI+DenoiseAI” should be used in different situations, after many experiments you will learn which is the best for each situation.

1 Like

AI Clear (or some facsimile of it) was merged into DeNoise AI as a processing option. It is not delivered for use in the same way as it was previously. See attached snip from DeNoise AI.

Fotomaker_DeNoiseAI-ProcessingOption

Figure this one out…

@jgr00ms2 Looking at the last three pictures I would choose the Studio AI Clear as the best and sharpest. I have posted that it was the best several times. However, as an observation, the sky noise looks more like jpeg artifacts instead of real noise to me. If it is artifacts that may explain some of the differences in performance of the various filters as they many not be using the same methods for noise reduction.

1 Like

Here’s a screen-shot @ 150% directly from the Raw processor:

Raw loaded directly into AIDN: (ORF)

Agreed on all your pts.

Have you tried a different photo program to see if it looks the same? RAW photos are very low contrast because no corrections are applied but your photo has corrections so it is being altered. It still does not look like normal noise. You can try bringing the RAW picture directly into Denoise AI, then after removing noise, do other adjustments to see if that makes any difference.

I’m not clear on your definition of “normal” noise. There is luminance noise and chrominance noise. Yes, I have several apps that can process the Raw file …no appreciable differences among them in this regard. It looks like color noise to me.

You may recall this thread regarding poor performance with color noise?

DeNoise-AI struggles with color noise

Including the statement about jpeg artifacts? Because the last image above is going from Raw->Tiff (no jpeg). Of course, I must convert to jpeg to post but I assure you that the Tiff looks the same. So, I struggle to see that jpeg has anything to do with it, since the noise doesn’t just appear after converting to jpeg format. ie jpeg has nothing to do with this issue.

This screen-shot of FastRawViewer looks the same as it does on my monitor (ie no jpeg involved)

Just for curiosity sake I downloaded the straight Raw snip to try with Imagenomic’s Noiseware.

See below for snip of result. It looks cleaner, to me, than the other posts above in which there is still noise in the sky (used that program’s “Landscape” preset with a few slider pulls to taste - both Luminosity and Color slider pulls seemed to help).

I took that Noiseware output and applied a light High Pass sharpening in Ps CC. Will attach the snip of that too. There’s still some fracturing around the railings and roof of the lighthouse. So wouldn’t want to blow up image large. But seems more forgiving in this size.

See what you think… FYI.

Noiseware (Landscape Preset) - Applied to Raw Snip Download:

High Pass Sharpening (in Ps CC) Applied to Image from Noiseware:
LitehouseFromTopazChat-Noiseware-LandscapePresetAndHighPass

I can remove the noise with DN-6. That’s sort of the point ie why DeNoise-AI cannot…or doesn’t.

1 Like

The reason I said the noise in the sky looked like jpeg artifacts is that it is splotchy like pixels. There is no color noise which often goes hand in hand with luminance noise. The RAW image processed in DAI looks more like luminance noise (speckled) and it was not improved much. Denoise 6 and Denoise AI use totally different methods so perhaps the AI doesn’t recognize the noise type in your photo. Denoise AI does an excellent job on most photos but there is always room for improvement and my earlier post found some artifacts produced by this program.

When you use a manual denoise program you just adjust until the noise disappears but you also lose sharpness which can be seen in the Noiseware shot before applying additional sharpening. Thanks for all the effort you did showing the results and comparisons. In a different post I showed a sky image (Hoover Dam) where the result was good using DAI but less detail loss with the AI Clear setting.

1 Like

My photos must lie in the category that hasn’t been trained for as I’ve yet to find one where there was a significant improvement over my previous tools, but I don’t typically shoot high ISO. And in the DPR thread the majority, including yourself, picked DxO Prime as doing the better job in that high ISO test case even though it’s not AI-based.

Being as AIDN was a free upgrade for me, I’d love to have it work well, but I have much better success with Studio:AI-Clear and DN-6. Yes, it’s obvious that the AI apps use new techniques over the old, but the overall end goal is more improvement which I’m typically not seeing, unfortunately. The fact that Sharpen-AI removed more noise than DeNoise-AI is still puzzling to me.

A post was merged into an existing topic: Adjust AI Color Shift

Post moved to the color shift thread for Adjust AI as this thread is about DeNoise AI/AI Clear.

Upon closer inspection, it appears something deeper must be going on under the hood. This specific system is not getting the demonstrated results on the TL test images either. Maybe that’s why I’m not seeing the typical results on any images. I have added this info to the support ticket.

1 Like

Support determined that v1.1.1 is broken on my two Win-10 systems (AMD GPU & Intel iGPU)…a furnished dev version (v1.1.2t) is working. So, the issue is internal to DeNoise-AI.

1 Like

Recently I had to shoot with a high ISO in low light. As expected, I got lots of noise. I ran photos through both DeNoise AI and AI Clear. Hands down, AI Clear was much much better. Please continue to support and upgrade AI Clear! Well, that’s my vote anyway

3 Likes