Used the generic mouse-click option. As others noted, the various subject category options are tightly trained as a tradeoff for speed/size. Personally I will take that tradeoff to keep the model local rather than “cloud-based”.
To wander off and speculate…
Broadly, this technology is still new and presume that DxO’s models will progress too.
Only checked a limited number of “problem” selections yesterday, but found PL’s implementation at least as good as LR’s. Count me impressed and pleased.
Note: I’m seeing excellent results running on a M4max chip with 48GB memory. Like the denoising technology, perhaps there is a more critical machine capacity for these AI models.
Addendum: Selection failed on some images C1 handled well and the mask was poorly defined in some cases. Let’s see what the next 0.x update brings once they get feedback.
I had to search a lot to find mention of this requirement, but I do see it now!
No, I’m running the latest drivers (and I’m reticent to downgrade them for one program - but appreciate that’s my choice right now).
I hope they manage to fix the issue so that the functionality works regardless of drivers. We can’t stay on increasingly outdated drivers indefinitely.
I have now heard back from Support regarding my issue with the Ai masking failing to work behind trees for sky selection and failing to recognise structures / buildings as subjects both via the ai dropdown menu.
This apparently is expected behaviour and Support says: - This behaviour is a current limitation of the AI masking feature. The algorithm is optimized for clear subjects (like people, animals, or simple horizons), and complex structures such as tree branches, fine foliage, or architectural details can cause gaps where the mask does not apply.
For now, you can refine the results by:
Using Control Lines or Control Points to expand the mask into areas the AI didn’t detect.
Combining Auto Mask with a brush to paint in the sky behind trees.
Adjusting feathering and tolerance to achieve a smoother transition.
Seems to me there is little point in upgrading from version 8 as I can mask just as easily using the old ways. There is no threat there to Adobe yet and I am inclined to agree with an earlier comment - PL9 is not ready for prime time. It seems like a rushed release.
As much as I’ve not had luck getting it working for me (I’m running recent nVidia drivers), I can see how AI masking can make some workflows quicker/easier (e.g. masking a person.
Is it worth paying for a new version, just for that? Well…!
This is a key point. Given how we’ve seen DxO continually refine DeepPRIME, I have high hopes for the AI masking capabilities. I think their first attempt is to be commended. It’s 85% of the way there, out of the gate.
But to release it as is is a joke. Why would I want to upgrade to that? No matter which way one looks at it, the release is not ready for prime time and the Ai masking is far behind the competition… I agree it will improve over time but how long will that take I wonder?
I have got onto support. Dell support say on my laptop (and others) not to use version 5 drivers as they are known to cause problems with drivers updated since they were used. I also had on my tower an video error from using Photo Supreme cataloger compressing. Its appalling to put out a program needing to install old drivers that in some cases are thought to be incomparable with latter drivers
I really like this first effort of AI masking. But it needs some more tweaking. On my system (Windows 11 i7 14, 4070 GTX) I don’t run into crashes etc.
But, the sky masking is nog very usable yet. Even in a blue sky with some clouds, it selects the sky, but with a gradiant in it. And also the earlier mentioned problem with sky behind trees makes slecting the sky more work than needed. When combining masks to fix this, it’s still hard to get the whole sky selected.
What I really mis, is the option to expand/shrink the mask and blur edges. That would make the masking options much more usable.
I have to admit that after playing around with the Ai masking for a while it can be useful…once the idiosyncrasies have been worked out. It is clearly a work in progress but it can add some value to one’s workflow. Whether it is worth the upgrade price at the moment is a personal thing. If they sort the bugs quickly then all is good imo - if we have to wait for V10 then not so good.