Topaz Photo AI v1.3.0

Topaz has released Photo AI v1.3.0. The highlight is the new Strong Sharpen model. Topaz said it "should have less artifacts and greater slider sensitivity." There are also other fixes and changes that you may find useful.

Thanks for this Annie, and great to see you here after our DPR days. I often find that Photo AI sharpens more than I would like. So in that case, should I use Strong Sharpen model, because of it's greater slider sensitivity? Thanks.

I've tried Photo AI 1.3.0, and strongly advise against installing it. Topaz has tried to simplify the UI, and no longer allows you to choose between motion and lens blur sharpening. It tries to do it itself, and seems to do it badly. Here's an example of an image with motion blur:

Now the image sharpened in Sharpen AI:

And in Photo AI 1.3.0:

I think the Sharpen AI version is noticeably better.

Many others have complained vociferously about this change; some have reverted to an older version, others are waiting for Topaz to sort out this stupid mess.

Wow, that is bad - artifact heaven with those blurred railings. Thanks for the heads-up.

Edit: I just noticed that the images you directly inserted into the text are all the DxO comp file. Should probably adjust the URLs to avoid confusion for folks who don't click at the links down below right away, like I did.

In each case, the starting point was the TIFF image file exported from PL6. I uploaded a JPEGs of less than 4MB from this, as well as the two Topaz outputs, again compressed to fit within the 4MB limit. I'm not sure which versions got displayed in the text (I'm not used to inserting images in this site). Hopefully it's not corrected.

Hello I am glad to see you here too.

Yes I also find that Sharpen is usually stronger than I would like. I usually preview the effect in the Topaz preview Window. If the image gets artifacts, I would not process further. My photos, which are mostly cat images, suffers more severely to Sharpen artifacts because of the texture of cat furs is not exactly a good target for sharpening. I suggest you to start with the AutoPilot suggestions and you should usually try to reduce the strength of Sharpen to see if it improves quality.

Hi Nigel, I am happy to see you post here :)

It seems the effect of the new Strong Sharpen model depends on the textures. For my cat images, the artifacts are less severe but still noticeable, especially at the edges of the furs or the eyes of cats. Also I believe this is just the first v1.3 series, so there would be a lot of rooms for improvements.

Oh yes, I'm sure there will be a succession of improved versions. I will suspend my use of Photo AI till this problem is claimed to have been addressed.

I was trying to illustrate my point of "textures" by re-installing v1.2.10 and compare with v1.3.0 and have a 100% crop of the outputs.

v1.2.10

v1.3.0

We can see that v1.3.0 produces a smoother output compared to v1.2.10, however artifacts are still visible on the edges of transition from brown fur to white fur.

Yes, in this case, TPAI 1.3.0 seems much better.

You got them under 4MB! But I think there is posting error here. Sharpen AI and Photo AI results looked similar, so I saved the files and ran cmp (compare). The files are identical. I like the image and decided it needed a gradient to reduce glare/flare in the left side sky. Sorry about sloppy feathering.

I think the three uploaded images are different, but in my initial post, the same image was erroneously used for all the inline images (now fixed). I'm still getting to grips with how this site works.

Thank you for reposting, Nigel. The two images are different size. I can see that Photo AI does not improve the camera-shake blur probably because, as you surmised, it is using the wrong AI model.

It's kind of nice that you can go back and edit your original post, a novel feature compared to DPreview.

I have an update to report — I might have been a bit too hard on TPAI 1.3.0, but it's still in the dog-house.

I later realised that it had defaulted to using the Standard model, not the new Strong model. What's odd is that the Autopilot decided it needed NR (it didn't), but not sharpening (it did). It should be smarter than that by now.

When I overrode that default, it chose the weaker Standard sharpening model, which is what I've uploaded so far. I've now done some tests with the Strong model (which includes the old lens and motion blur models), and discovered that it defaulted to a very low setting (8). I then experimented with various stronger settings, which definitely improved the results. But something odd happened when I exported an image with the setting up to 50: it looked fine inside Photo Ai, but the exported image had a strange bloom. I suppose it's an artefact, but not like the usual ones.

So, my conclusions are that, if you work hard at it, and override all the default Autopilot settings, it's possible to get good results as good as Sharpen AI produces fully automatically. But that seems like a serious step backwards. So, I'll stick with Sharpen AI for the moment.

First, here's the export from Photo AI 1.3.0 with the Strong model with the low strength setting of 8 chosen by the Autopilot:

I think that's much too weak — it seems to be bending over backwards to avoid artefacts by doing very little sharpening.

Here's the output using the new Strong model with a much higher setting of 40. I think it's pretty good:

And now a setting of 50. Note the strange bloom towards the left, which doesn't appear in the preview. It makes this rendering unacceptable:

In contrast, this is what I got, with no need for any experimentation, from Sharpen AI 4.1:

So Photo AI still has a long way to go. But I'm disappointed that Topaz still appears to do so little testing before churning out these weekly releases.

My impression too. It seems like a tradition for Topaz to rush out new features in ".0" releases and refine in later minor updates.