The “ghost” duplicates did not appear at 50% or less screen magnification. Disabling and enabling Perspective tool fixed the problem, jpeg exports were not affected even when the problem appeared on the screen. Changing Perspective parameters made the problem appear again. AI mask was used to select the dark subject. Maybe the issue appeared because I used Perspective tool AFTER using AI mask (TBC)?
The photo required some very heavy shadow recovery and highlights suppression - the sky was “dirty” but bright, it was a bit rainy, the subject was in the shadow. The problem seems to come only in special situations. I made about 100 similar corrections using PL10.0.1, and I’ve noticed the issue only today.
Config: PL10.0.1, EIZO CS 4k 27" monitor at native 3840x2160 resolution, NVIDIA RTX4070 with 596.36 studio driver, i7-14700KF (no iGPU), 32 GB RAM, Nikon Z8 14-bit RAWs (45mpx), Win 11 25H2 (26200.9457), System->Display->Scale set at 175%.
I also get “excessive” PhotoLab.exe memory usage from time to time (more than 12GB, while usually it’s 4-6GB), which looks very dynamically managed. My initial suspect was also Perspective tool, but I didn’t have time to look into it – maybe next week.
UPDATE: I think the problem appeared because I applied Perspective tool and Crop AFTER applying AI Area mask. In all my previous work AI masks were created only after Perspective/Horizon/Crop tools were applied, so the problem did not appear. Sorry for possible confusion.
mwsilvers
(Mark - New Jersey, USA, Win10, RTX 4060, PL9, FP8, VP5, Nikon Zf )
2
While your UPDATE may have resolved the issue for you, in addition I noticed that your Nvidia driver is almost 5 months old. It has been updated three, or perhaps four, times since your current version from late April. The latest studio driver is 616.92 released on September 9, 2026.
I don’t want to update NVIDIA driver before I run into specific driver compatibility problem. I’m not one of those who followed the general guideline and upgraded to 616.5 early6, hitting new problems. From the performance perspective, NVIDIA drivers are mature enough to have all major performance issues fixed (mostly sometime in 2023, btw Intel seems to do it now), except possibly for the new TensorRT-RTX Execution Provider (but that’s partially taken care of by DxO). From the stability perspective, I’ve seen enough problems with NVIDIA drivers (just a few from my experience – flickering, false GPU temperature reporting, power management still not fixed, TDR crashes – even with Win Notepad!), and I’m not playing any games, I don’t use G-Sync, HDR, multiple monitors, or any other “advanced” features". So, as long as I don’t get any application or kernel errors in EventLogs, and I don’t get motivated guidance to update the driver, I’ll not do it. My top guidance is: “If it works, why fix it?”. In case of my problem, I don’t think it has anything to do with the NVIDIA driver – these have other type of footprints.
I think the interaction of PL “picker tools” with “geometry tools” needs some clarification from DxO, perhaps in a form of FAQ.
By “picker tools” here I mean HSL, SmartLighting in SpotWeighted mode, WhiteBalance, ToneCurve, Local Adjustments masks, maybe “Set Center” of Creative Vignetting (did I miss something?). By “geometry tools” I mean Distortion, Horizon, Perspective, Volume Deformation, ReShape Fusion, Crop (again, did I miss something?). Crop tool might be some special case here, as it probably uses some absolute coordinates under the hood (?).
Why the “ghosts” appeared only at > 50% is a bit of mystery to me, seems like a result of PL bug. I will experiment, try to reproduce and take a closer look on Monday, before reporting to DxO.
Partially off-topic:
Fully documenting the processing pipeline in detail, the sequence of correction steps and coordinate system used in each step would be welcomed. I have little hope DxO would do that, but they should at least provide some practical guidance. For example, it seems to me that PL processing pipeline partially looks as follows: Channel Mixer > Rendering > LUT > Tone Curve > HSL. So e.g. you can’t use HSL for B&W rendering, and you are left only with different and less precise Channel Mixer settings – an example of practical knowledge coming from this. What about other corrections?
There’s a lot in this article. More detail about when certain adjustments are applied in the pipeline might be found through searching the forum or combing through DxO’s website - but I’m not sure the reward is worth all the work. For example, I remember coming across a comment about when LUTs are applied - but am not able to find it now.
Sure, I’ve seen the published DxO PL pipeline description, but it’s mostly marketing, an overview with little practical value. On the other hand, I’m aware of scientific research done by DxO and its coworkers, and I may wonder why it’s not mentioned there (Morel, Bruyn, …, iirc). But it’s their business, not mine. My point was very practical – give us enough knowledge of PL processing pipeline, so that we can adapt it to our working needs. That would mean in particular explaining in “practical detail” the interactions between “geometry” and “picker” tools.
If the process provides insights into the development model, DXO will surely become very tight-lipped.
I once tried to understand how Microsoft Word is programmed and then gave up. There are too many possible variations and perspectives. That could also be the case here.
Then I still have in mind (don’t remember where I got that information from), that DxO applies the nondestructive (or all ?) corrections in “its” own specific order. This means you can use the tools in any order … with the exception of those that result in geometric changes.
Perform lens corrections first, then crop then all other adjustments and lastly local adjustments, including healing and cloning, once you are satisfied with your crop and rotation.
EDIT: It is the rotation that causes issues with LAs and clone/heal tools.
Unfortunately I could not reproduce my problem. Looks like some race condition, maybe I was too quick with something.
Previously the “ghosts” appeared only at > 50% and they looked like exposure correction applied to the mask as it looked before perspective tool was applied. The single mask consisted of two submasks, both created using AI Area selection. Just tried it again, and the “ghosts” no longer appear. Although the submask gizmos are somewhat displaced but that does not affect the image as shown at any magnification or exported, the submasks are displayed correctly. Strange.
@Wolfgang – I’m aware of that DxO FAQ article. In this case AI Area selection was made, so it probably doesn’t matter too much and it worked correctly with UI at 50% or less and on exports. Anyway, the original problem is not an issue, if masks are applied after Distortion, Horizon or Perspective, i.e. in the correct order.
Yes I often do different crops on a same image, and as I said, I never noticed different crops affect my result, so I wanted to know if I missed something. Are there some combinations which lead to have different results when crop is done last ?
Maybe it’s only about Distortion, Horizon and Perspective (maybe rotation too)?
The Crop coordinates could be simpler to handle.
DxO FAQ description, as it currently stands, might be NOT precise enough.
I’ll try to investigate it next week, unless another Out-of-Seq job drops in…
Do the curve tool apply its tranformation before or after the selective tone and so on.
Indeed we can spend time to test all this with color and black and white charts as image. But users should not have to do some kind of reverse engeenering and shoot raw charts to understand their software !!!
Probably you mean Horizon and Perspective. Distortion may also come into play (?).
I’ll try to make some tests, but probably someone has done that already?
So the question is: Does cropping alone have any impact on AI masks?
