The new Normal V2 Beta is WAAAAAY better, and I wish I could default it somehow in setup to save time changing it. No more halo’s, etc. Also WAAAY better and not creating blobs in background bokeh, and leaves it smooth. yay! It still does this slight thinning of narrow objects, like a butterfly antenna ends up looking ever so slightly narrower, but maybe that is part of the halo reduction technique? No idea, but even with the slight narrowing, it looks way better than a white a halo around the narrow object. I still wish there was a “de-sparkle” button, heh heh, to get rid of the white dust on black objects getting so much brighter. I end up using the de-select tool to get the program to ignore those areas, so as not to emphasize the sparkles as I call them.
Agree, eg all details/structures in a harvested field are totally blurred with normal v2 while fine structues remain when using normal v1! After switching a few times between v1 and v2 the app crashes without error message… After restart of the app this finding is not repeatable. Suddenly normal v2 performs visually much better, nearly as good as v1. Strange…
Tested the new model on a ruby-throated hummingbird in a tightly cropped image and it was considerably better at both noise removal and sharpening than the original model. There was moderate noise in the image, but not severe. No blurriness. I did find it useful to use the slider for both changes rather than auto-parameter. For noise removal slider was set at 9, detail at 10. For sharpen, strength at 10, clarity at 7. These seemed better on-screen than auto.
Super results playing with a few really bad images, as good enough images do not really give a definitive answer. My absolute support for changing to both version 2’s. They are without a doubt a remarkable improvement in discerning for example background and especially challenging lace material but not so good on fake leopard blanket which didn’t look like a good likeness anyway. In the last case some spots were well defined, and some other places had weird ringy lines where previous industry efforts more likely resorted to smoothing out detail into flat look. However most amazing.
Overall I’ve been liking the new v2 models on pictures of people, especially in a dark environment or with a lot of noise (scanning older negatives).
One area I’m finding trouble with the new model is on more technical photos, like this airplane. I was really happy with the results I got on 1.3.3 but unfortunately I can’t find exactly which settings I used (put this in the metadata!). I tried to recreate a comparison here.
Notice the rivets in the plane window on the 1.3.3 version, missing in the new models because they are smoothed out as noise.
Also the “Dobbs” truck in the background gets overly corrected and the details look bad.
I typically use the "“Detail” setting really low because it looks better to have a bit of blur rather than overly sharpened. I can’t find a balance here though.
I like the Normal v2 but generally keep the noise strength very low at like 5, anything above that makes things look very plasticy and odd, but around 5 it is great. However my main issue with Normal v2 is that it crashes…like all the time…win 11.
Looks like the V2 beta normal noise reduction is much better for a milky way photo from last night
Yes at very low settings they (both beta models) are better than the old models at this kind of images and better than the other vendors A.I. denoising models bc
TPAI is more adaptive.
I mainly use TAI for improving digitized 30-40 years old slides (heavy noise). In my opinion Strong v2 is much more better than v1. The look of the images is not as “plastic” as before and the slider is much more sensitive. This gives you a better choice, how much noise you want to remove.
The beta version (Normal) works better (for me) on portraits in group shots. A ‘plastic’ looking skin can be avoided completely. Definitely an improvement.
V2 produced a more pleasing result from a very noisy Astrophoto of North American Nebula.
It eliminated the distracting background, yet extracted numerous stars in the foreground.
The result looked good.
I am post processing pictures from a concert. So very noisy images since I was using very high ISO. As far as I experienced, the V2 is always very very noisy and blurry, while V1 is more effective.
I have just starting using this beta v2. Just brilliant! Thank you, hope it’s a permanent feature from now on.
It looks great, but it truly always crashes after making 1-3 adjustments (3950X / RTX 3090 / 64 GB).
Just adding a screenshot to how pleased I am with the v2 beta normal. Detail within this image of Lincoln Cathedral, UK is just awesome.
Well, for me, sometimes the version 2 (beta) Normal Noise Reduction is better than Normal Noise Reduction, and sometimes it isn’t. Haloing can be a problem in one version, and not the other. I suggest keeping both version 2 (beta) and the normal noise reduction. I
I’m still hoping for “Weak” options for both Noise Reduction and Sharpening.
Alan Woods
Oregon
V2 beta is nowhere near as good as the original at removing lithographic artefects from scans of published photos. Can we keep the original? (I haven’t played with the sliders, so that may help).
Peter Walker
When processing a photo both with Topaz PAI Remove Noise Normal and with then with Remove Noise Normal V2 (Beta), I see a significant difference.
The existing Remove Noise Normal will create some white fringing, especially along dark to light transitions. It may also create purple fringing. Remove Noise Normal V2 (Beta) does not create white or purple fringing.
So I’m spending a lot less time getting rid of the fringing problem.
Go ahead with Remove Noise Normal V2 (Beta).
Dan Laughman
The 1.3.7 release has broken both Normal V2 and Strong v2 producing the error “Could Not Run Model”.
RTX2070 Super with 531.68 Nvidia drivers on Win 11.
I had to downgrade to Studio v528.49, try that and see if it helps. My 3060 with the latest driver had issues with a couple of apps, not just Photo AI.
Note for the processor select Auto rather than the specific GPU
