Normal and Strong v2 Model Discussion

We’ve made updates to both the Normal and Strong Remove noise models. These models now cover a wider range of noise amounts than before. The Normal model is suitable for small to medium amounts of noise, while the Strong model can handle any amount of noise. Our goal is to use the Normal model for regular noisy images where preserving details is crucial. On the other hand, the Strong model is designed to remove significant amounts of noise, and slight blurriness may be acceptable at higher slider values.

These beta models should be more responsive to the slider’s value. Both beta models retain some noise at smaller slider values, and gradually denoised the image based on the slider strength. Therefore, a good auto-parameter setting is very crucial for these models. We’ve also fine-tuned the autopilot setting for the new denoised models to maintain natural-looking results in most cases.

These new models are going to replace the released version of the Normal and Strong models in an upcoming release. We like to hear your feedback about the output image quality and about the effectiveness of the auto parameter setting.

6 Likes

Are you going to make the BETA Normal and Strong V2 Model available for processing a RAW image via “plug-in extras” in Lightroom soon?

2 Likes

Love Normal V2. Much more controllable. The old normal when slider at 0 would still denoise too much. With Normal V2 I finally have control to have a little noise if desired. Plus the details look great.

4 Likes

No, these models are for non-raw images only. We will also update our raw Denoise models, which you can use to process a RAW image via “plug-in extras” in Lightroom.

Any idea when the updated Denoise models that can be used with RAW files will be available? At least in some BETA form like the non-RAW ones.

1 Like

We are actively working on it but no solid deadline yet. You can see the features we are currently working on in our roadmap post.

Mac mini M1 (2020) Monterey 12.6.5

I like the results I am getting from Normal V2 (Beta) on TIFF files of birds. The sliders seem to me to allow a finer bit of control on both Strength and Detail. After comparing Normal to Normal V2 (Beta) started to use it. Thanks.

2 Likes

Both Normal and Strong v2 are too extreme, particularly what it does to faces

1 Like

Normal V2 beta crashes for me when used on tiff files from capture one when you switch from normal to v2 beta and adjust the strength. It reliably crashes with no error given, Photo AI just vanishes…
Ive never had a crash before with Photo AI or the other topaz tools.

After specifically testing and comparing the result on various pictures
The V2 Beta seem to produce more artefacts specifically in fine structures like trees or hair, thin branches look odd, thinner or partly disappear.
The appearance of the picture does change. At this point I will stick to the none beta for the majority of cases.

The positive improvement is how it takes on noise in small enclosed spaces, like a patch of sky visible in a tree between branches.
Where this is left almost untouched in the old model, it does a much better job in the v2 Beta.

Also in extreme noisy situations (say dark and ISO 25K) the v2 Beta delivers more crisp results, less noise and looks sharper.

2 Likes

Normal V2Beta worked very well in old scanned images in correcting the amount of noise. It was an improvement for old photos with lots of noise

I am still testing – and I have found two things:

  1. the autopilot detects and removes noise even if there is almost no noise with the result that the image is much more blurred and did not get significantly more details. When comparing this with Adobe Denoise AI, the result can only compete if the Remove Noise Normal V2 is turned off and a slight Sharpen effect is applied.
  2. There are some unwanted artifacts among fine details. I have seen similar things with the sharpener in wood structures. And they are back here too:

Just to be added: I have also tested the strong beta v2 model – and even increases the artifacts. So far, I cannot see any use for the strong model since it kills almost every detail. As mentioned already many times: pls bring back the grain option. It can cure



so many artifacts and plastic looks.

and here is a comparison between the Remove Noise Normal beta v2 from 1.3.4 (l) and Adobe Denoise (right). Pls note the details and the atmospheric “noise” in the image to the right – and the more artificial look (what I call plastic) in the Topaz variant to the left.

You can lower the denoising of beta normal until the noise comes back.

The image of TL is clearly more denoised as the one of Adobe, you can also raise adobes denoising to 100% and then compare this.

During beta i did compare 30%, 50% and 100%.

A gamechanger is to denoise the output of the other denoisers with the one of TL.

Thanks - I have tried this but I did not really reach the point where the noise comes back as I would like to see it. I have tried down to 0.

Sorry, must correct myself – I am reaching a zero level and the noise is there again. Thanks for the feedback. I am testing combinations of Remove Noise Normal v2 at 15/3 in combination with Strong sharpen (6 ). The results look quite good – and somewhat better to my eyes than with Standard Sharpen. It would be handy, if there would be an option for saving presets!

1 Like

Photo AI 1.3.4 crash on Windows 11. Having tried jpeg, tiff & raw files (cr3 & rw2) on the new denoise models, the strong denoise model worked fine for the raw files but with both jpeg & tiff strong denoise classic crashed the app. I have uploaded the crash log.
2023-05-15-13-39-30.tzlog (135.5 KB)
The strong V2 model seems to work ok on jpegs. Hopefully this will be fixed soon.

Normal v2…the new norm. I like it. So far appears to provide more “natural” results and not look over processed.

2 Likes

You can use Enhance, i suggest standard, at the same time too to get rid of remaining noise and get more detail.

But this does increase the processing time by a lot.

I does use this very often to finetune my images.

You can also increase the size of an image and downsize it again to get an even better effect.