Just had an email from DXO support to say NVidia has released new driver 581.5 which fixes AI Mask problems. Internal tests confirm fixes. Will install tomorrow. Anyone had info and tried.
Ken
For my setup with laptop RTX4070 use of predefined masks (like Sky AI) still causes error when exporting to TIF.
@kena some other users think it is fixed but I donāt New Nvidia driver 581.57 with PL9 AI Mask fix! - #23 by BHAYT
More detail on this here ā¦
NOPE.
Nvidia Studio Driver 581.57 on RTX 3090.
Initially, the new PL version appeared to work (e.g., āselect backgroundā).
Beware: Now, after trying to open the images I applied the āselect backgroundā to, I get an āInternal error (Correction failed on the Execute stage)ā.
I cannot fix this - cannot edit the DNG because of that fatal error - workaround may be to delete the .dop file to erase all edits.
Not good?
Your changes/edits will be stored in the database too - Youāll probably need to rename the file to force PL to consider it a ānewly encounteredā image.
Well Iāve just installed driver 581.57 and tried AI masking. Each image I tried failed with the same old fault. Shut down and then re-opened PL9. Each image I tried with AI masks worked fine. Very strange. Shut down and re opened PL9 and AI masks failed again. Shut down and re-opend masks worked. This is not good and Iām going back to to 576.40
Iāve gone back to 576.40 and all seems to be working OK. Will leave here until there is any more reliable updates. using this driver does not seem to affect anything else on the PC. I will be getting back to DXO with my findings.
Very strange that for some of us the new driver improved matters, for others made it worse.
In my case AI masking works better now, but I did encounter a new export error earlier and Iām damned if I know why. It was the same error that the usual AI masking failure resulted in (according to the error log). Restarting stopped it happening again (so far).
Our job as beta testers continuesā¦
@Fineus Shhhhh one of the rules as a Beta Tester is that you cannot admit or deny you are a Beta Tester! ![]()
Looks like the testing continues as you say
I was wondering about that. Once an image has failed with the error it may be the reason it continues to fail.
@KenA If it fails with
then a program restart is mandatory, in my opinion.
If you encounter an export error then typically all subsequent exports will also encounter problems. Whether terminating the export worker at that point might reset things I am not sure but if the exports are going to fail is there any point in continuing with that run!?
However, I have encountered problems such that I need (or felt I needed) to restart the machine and those are rather odd because continuing with the old database and DOP has worked on some occasions, so what ācorruptionā was causing the restarted program to fail as quickly as it seemed to?
The only sources of corrupted data are the database and the DOP so the most extreme is to clear the database before continuing, and I have done that but mostly as a precaution not because I felt it was a necessity and/or removing the DOP but that means starting the whole edit afresh.
I have not resorted to deleting the DOPā¦yet.
Plus is it even vaguely possible that Photo|Lab is corrupting the running Nvidia driver environment, hence necessitating a full machine restart!?
Perhaps try moving the image and DOP files to a new folder.
Thats what. I found with a small test on a laptop
So tonight I deleted the folder of images I was testing and reinstalled using a different file names. I then reinstalled driver 581.57 and found that the keyword masking seemed to be working OK. Although, I did have one case where it could not find the subject and then the error occurred. I shut down and restared the PC then had another go at the image and it seemed to work. Iāve tried other folders of images and had more success. I think for the moment I will leave the latest driver 581.57 and see how things go. Clearly there is still a stability problem as far as Iām concerned.
Just had new update of NVidia game driver advised, 581.80. Decided to leave at the moment as PL9 masks still working ok with 581.57, unless anyone else has tried it.
@KenA Your post duplicates this earlier one:
