For brush as all other non AI thing I think (guess) its a normal way. Its give more easy way to store editings, without this undo can be very problematic, layering changes (order) can be possible, masking tool ‘upgrades’ updates can be works, etc.
For AI mask, as AI calculate, the good prompt is the ‘only’ real parameter, and off course all the previous thigs van be valid.
I think (guess) use polygons can be problematic. At general: for example handle vectors for pixel accuracy. Feathering can be an issue. In the case of new/upgraded AI model all vector need to be updated. Etc.
It would solve all performance issues
Or may performance degradation expected. I’m a bit unsure it’s really helps. Other softwares when polygons use, still do a lot of calculations for pixel level results, re-do edge detections, etc (at least based on my experience). May accurate poligon calculation still takes a same resources.
I think ‘pixel level flattening of all mask’ may can helps (in final step before Exporting), but i think its may raise issues in Noise Reduction, made software development more complex. Note: AFAIK also LR not does mask flattening (and its may one of the more requested feature for ages in LR?). I guess for some same reason as PL not do it.
Other: few AI part in DOP in the meantime: Subject, hair, Single mouse click, Area selection
I have DxO PL 5 (because I skipped v6 and v7, and I was in no position to be able to afford the full price for PL version 8). Now, for what I have been reading regarding version 9, I won’t be able to upgrade also for technical reasons, as I have a notebook with an integrated graphic card with very low GPU memory. It is not in my plans to buy a new computer either.
I also have ACDSee 2024 which has some AI masking utilities. It works fine in my computer. The program (up to that version) depends mostly on CPU and AI masking is done in a breeze. But then came version 2025 for which AI functions became to depend on GPU. No only the added AI masking functions to the “Develop” mode are impossibly slow in my equipment, but the processes that used AI that were already existing in version 2024, became also very slow (so much, that I will not upgrade the program). It seems that the changes for the newer version are beneficial for those having good graphic cards, but not for people like me, having an older or less powerful equipment.
I also have Affinity Photo, and it has some AI masking functions that work fine with my notebook. I assume that the program automatically and efficiently chooses how to balance the use of CPU and GPU according to the characteristics of the computer being used (what seems not to be the case for PL v9 for what being said on this forum thread, nor for ACDSee v2025).
You have missed the obvious truth that PL9 was tested with the most current drivers and the observed issues could not be fully resolved before PL 9 was released. The result was DxO’s suggestion to use an earlier driver. Whether the release should have been held off until the driver issue was fully resolved is a separate question.
The PhotoLibrary has never really worked right for me, going back to PL5 (when I first drank the PL cool-aide). Often, when you add more or remove photos to an existing folder, PL chokes and refuses to understand there are new photos. When you remove photos, the filmstrip at the bottom just shows generic question mark icons. Given that we are now at version 9, this will never get addressed.
I can’t argue with any of what you are saying. Whether it will be very costly to them is hard to say. In the past a number of their competitors have implemented buggy releases that infuriated users until issues were resolved, but no one seems to remember those issues now. The original introduction of the Generative AI feature in ON1 Photo Raw was one example. A second example was a version of Luminar from a few years ago that was very buggy although I don’t recall the specific issues.
User communities are a fickle group of people. Some will condemn software over one specific missing or broken feature that is important to them and others may not care about that feature at all. They will also change from condemnation to praise and back again in a heartbeat.
I will likely only use the currently slow and buggy AI mask feature occasionally even if it was working exactly as expected. I certainly agree that having to download an earlier version of a graphics driver to fix bugs and achieve expected export times is not a good policy. Hopefully, the driver issue will be resolved soon. However, for what I need to accomplish most of the time, PL 9 works just fine as is with the older driver installed. I can take full advantage of almost all of the many new documented and undocumented features with no bugs or performance downside at all.
With my RTX 4060 graphics card and the older driver my exports of the 24.5 megapixel raw files from my Nikon Z f, with no local adjustments, are very fast . A single DP 3 export takes 6 seconds. A single XD2s export takes 8 seconds. Exporting the same images with the addition of a few local adjustments, other than the AI mask, is a bit slower . With the LA adjustments a single DP 3 export takes 10 seconds and a single XD2s export takes 14 seconds. I certainly can’t complain about that.
There is always a code freeze date when the GM is reached and all is prepared for release.
NVIDIA have their own release schedule and even if the latest nvidia driver released was used during the last test for PL9 - it does not guarantee that there will be another “latest” driver available at the time when PL9 is made publicly available.
I have no idea how the dates align for PL9 final and 580 drivers but it can be as simple as that which cause havoc for so many.
That is not correct. It suggests they did not test with the latest drivers and that the testers never tested with the latest drivers. The recommendation to use an older driver was a obvious result of their inability to fix the issues caused by the latest drivers within their release timeframe.
That is a poor and inaccurate assumption that suggests DxO incompetently developed PL9 with old drivers and belatedly tested it with current ones making it seem like their development protocols need to improve.
Development and testing of features takes many months and in some cases much longer. DxO uses updated drivers as they become available to them. Driver 572.83 which works just fine was released on March 18, 2025. Nvidia releases new drivers every month or so and as a result the first driver that would cause any issues was probably released towards the end of April or the end of May after much of the development effort had already been completed. The timing was unfortunate, but to suggest that DxO’s development and testing procedures need improvement and the current driver issues were a direct result of development and testing lapses is unfair.
I was specifically responding to your comments regarding the driver issue. I am using an RTX 4060 graphics card. With the older driver installed on my Windows 10 machine, other than known bugs and performance issues with the new AI mask feature, I am not having any other significant problems with PL 9. I have had no crashes and no internal errors of any kind and exports are now as fast as I expect them to be since changing the driver.
While PL9 does introduce the largest number of new features and improvements to date, I don’t think “by far” and “three times the average” are accurate comparisons to previous releases. I’ve been tracking major releases and their differences ever since Optics Pro 9. While some major releases in the first half of that period only brought a few new features, these still tended to be significant. Bigger leaps started with PL4, IMO, and have been steadily increasing since. Very exciting to consider that this trend could continue, given the size of the requested-feature backlog. But as you’ve pointed out, the downside is greater pressure to test for and fix bugs. There’s also DxO’s track record of only modestly implementing new features and then moving on to the next thing without further refinements.
With all due respect we may have to disagree a bit on this one. I too have tracked all the changes to PhotoLab starting, in my case, with PL 1. It may have been a bit of hyperbole on my part to suggest three times the number of new and updated documented and undocumented features. However the actual number still far exceeds double the average number. Would you be happier if I suggested two and a half times?
Are you including the undocumented feature updates such as eliminating the requirement to zoom to 75% to see the results of the Lens Sharpness Optimization tool and the Unsharp mask, and the update to the Chromatic Aberration tool lowering the zoom percentage to 50% from 75%? Or the new License management feature not mentioned in the release notes? Or the undocumented significant improvement in Auto Brush masking?
In my most recent very quick and rough count I found well over 20 discrete things that are different in PL9 and I believe the number is actually higher. I will attempt to itemize the exact number and post them here. There is an unprecedented number of new and updated features in PL 9 and given that, it is not surprising that there are some annoying issues which need to be resolved.
I don’t see the specifics as worth arguing about, even in private. Previous releases also had undocumented improvements that were simply grouped together as unspecified improvements in the UI, in performance, or in functional effectiveness. If you count them all individually, they can add up to a lot. But I wouldn’t give them equal weight and I’m guessing you wouldn’t either.
Is the ability to manage license activations limited only to PL9? If so, that’s more than a little disappointing. In fact, that would make this useless for migrations involving multiple DxO products for some time to come.
The image preview improvements in PL9 are great, yes - but small in comparison to adding full X-Trans support (carried over from PL8) or new mask options. And like many other improvements a fair number of them are only half-baked: yes, better than before - but leaving lots of room for further improvement. Don’t get me wrong - I’m impressed with PL9. But I’m also looking past the hype and critically considering if it’s worth an upgrade. For me, this is the first time I’m saying no - not yet, anyway.
I agree that normally I might not give undocumented feature updates the same weight as new features but in the case of PL9 I do.
Eliminating the 75% zoom requirement to see the results of sharpening at any zoom level is a major improvement for many of us, including me.
You are on this site almost every day and I assume have also seen how many recent posts there have been complaining about the lack of license control by users. For those folks the addition of that feature in PhotoLab might have major implications .
And finally, after all these years, the auto brush masking upgrade has finally turned a mediocre tool into a very useful one.
These may have been undocumented upgrades but to my mind they are not minor ones. They are all very important and very valuable upgrades and I give them equal or greater weight than many of the documented enhancements.
Mark
2 Likes
Stenis
(Sten-Åke Sändh (Sony, Win 11, PL 6, CO 16, PM Plus 6, XnView))
119
I will test a little more with 572.83 from March 18 before I decide to trust it fully out. Version 572.83 is the version 17 versions back. Nvidia has a really high release reneval speel for drivers. 17 versions is not even under a full half year.
Photolab 9 actually seems to work fine now with this old driver. I have no problem with the export-speed really. It is faster than before with Deep Prime as long as I don’t add any masks. With a typical set of masks it takes 25-30 seconds.
I tested even the latest driver from August 28 (581.15) but it was just terrible. It crashed even when I tried to add just a single “Person”-mask, so it was totally useless.
Even beyond the Nvidia driver issue PL9 still requires additional performance tweaking which I believe DxO is actively working on.
Mark
1 Like
Stenis
(Sten-Åke Sändh (Sony, Win 11, PL 6, CO 16, PM Plus 6, XnView))
121
Why is that? Anyone at DXO on Windows would have been able to see these problems at least as soon as exporting or printing at least. It is really strange that they have missed these problems. It is more understandable reviewers have done it because they might either export or print when reviewing but no way the developers have missed this. Maybe they just hoped Nvidia would fix the drivers before the release date - but tough luck, they didn´t and here we are.
If DXO shall have the slightest luck with attracting dissatisfied Lightroom users with this really feature packed version they can´t handle a release like they have and especially not
