This week has a host of bug fixes and improvements surrounding metadata handling and crop. For those of you on Windows who have run into metadata suddenly missing on files that contained paths with non-ASCII characters, that issue should now be resolved. There are also a whole host of crop issue fixes with more to come in.
Also as a small note, the simple non-AI masking brush is still a work in progress and is likely going to be in next week’s release barring any last minute issues. We know this is a highly requested feature, so just letting you all know that we are indeed working on it!
Please give us any feedback or report issues with this release. We’ll be updating TPAI regularly to address those pieces of feedback and issue reports. If there’s a specific image you’d like us to see, you can send it us at this dropbox link.
v1.2.3
Released February 23, 2023
Mac: Download
Windows: Download
Changes since 1.2.2:
Added compression slider on PNG exports
Added support for opening files with extensions jif, jfi, jfif, and jpe
Lightroom Plug-In should now copy Rating and Keyword metadata correctly
Fixed files with non-ASCII filenames and folder paths not saving metadata on Windows
– Anyone who turned on the beta UTF8 region setting as a workaround can safely disable it
Fixed issue with face refinement being blurry
Fixed issue where Enhance slider values were reset automatically
Fixed open image shortcut not being disabled correctly while in plugin mode
Fixed not being able to flip a custom crop ratio more than once
Fixed aspect ratio not changing the first time after unlocked custom aspect ratio was selected
Fixed crop not preserving correctly after flipping the ratio
Fixed crop aspect ratio going out of bounds in some instances
Fixed crop box moving erratically when dragging trackpad along sides
Fixed some issues with installer on Ventura not being able to write to the Home directory
Fixed issue where “dllhost” would show up as a host editor on Windows
Updated ExifTool to 12.56
Known issues:
If you drag the crop handles very fast into a corner, the opposite edge will change positions.
It is possible to get the crop ROI to be out of bounds in some circumstances. If this happens resetting the crop ROI or cancelling the crop should avoid a crash.
I’m getting a nasty tint from Lightroom → TPA → Lightroom. This started with 1.2.2 but I didn’t get around to logging the issue. Source is an ORF file from Olympus OM-1. I’m attaching the screen shots out of Lightroom.
Sharpening is not at all improved , I have tested same BEE image, one I send it to you, still sharpen whole . even I choose half, not impressed at all.
The last few releases have mostly been about bug fixes (well, also Apple Photos compatibility). Kudos to you and the other developers for doing this. Bug fixing means getting a stable product that works well and that is very important. Having a good, stable, consistently designed and behaving product will win hearts and minds whereas buggy, unreliable, inconsistent design, and annoying problems will do the opposite. Looking forward to the new non-AI brush, even better models for sharpening, denoising, etc. Good luck to all the Topaz staff in getting the work done with quality! Thank you.
Over the last few releases i noticed that face recovery has become soft in several situations. I am pleased to say that I have not noticed that issue anymore.
Great update…
Win 11 Pro desktop PC. PAI 1.2.3 Plugin. Processor = Auto (AMD RX6800 XT, Intel i9 12th gen).
This rel. installed & launched okay as Standalone & in Ps 2023 as plugin vers. Worked with Plugin for experience below…
This image makes the case for a feature I’ve previously requested (others upvoted with “Likes”)… a way to mask out (& invert or not) Denoised areas of images + ability to apply genuine zero (0) denoising.
PAI A.P. only turned on Denoise for this image. Not Sharpen. I added Sharpen when I didn’t get the appearance I wanted - to experiment.
In the BEFORE view (see labelled screen snip below) the detail in the buffalo’s eye is clearly visible - there’s a black area across the eye. Plus, the white details around the eyes (lids, etc.) aren’t too accentuated in that BEFORE view.
In the AFTER view (see labelled screen snip below), I couldn’t find a way to get the Denoised image to show the desired BEFORE eye detail (by trying various settings using denoise sliders only & various settings for denoise/sharpen) or a way to prevent the accentuated whts of lids, etc. around the eyes.
** NOTE: For animal, bird, etc. nature photography, sharpness of eyes & details in eyes are critical to capture/maintain.
This type of image is one in which I wish I could mask the denoised areas that cause loss of detail & then invert the mask (since, PAI masks/overlays seem to indicate where a given feature will be applied) so only the other areas of the image (not the eyes or snouts, for ex.) would be denoised.
This also would be a great place to use the manual masking brush that’s in development, b/c there are very selective areas of the image to be chosen for masking that a snaking, looking-for-edges AI Brush would over-run. For me, the overly-active AI Brush would be a timesaver (hopefully) to fill larger swaths of content. Not to make precise, delicate, limited area selections.
BEFORE view of Image (see notes above re: what to look at/for):
Hi Jayesh - we haven’t made any changes to our Sharpen models in the last few weeks so it’s expected for the results to be the same. Our research egneineers are working on a new Sharpen model and it shouldn’t take very long to be ready and implemented. For the bee image getting sharpened even when half of it is selected, we still can’t reproduce this with your image and we haven’t received other reports about it. If it’s not possible to record a video of the behaviour, could we schedule a screenshare so we can take look at it? Thank you!
Hello,
it’s absolutely stunning how Topaz Photo AI can improve the quality of blurry pictures (often scans from old slides).
But the difference in details of faces and their environment is often disturbing: the faces clear and precise, their environments (even skirts and trousers) still blurry. What can I do to force Photo AI to make everything detailed and precise?
Great to hear that, many thanks.
Once question though, does the masking only applies to sharpening or also to noise reduction ?
I hope it’s just for sharpening, cause NR is often desirable on the whole image but sharpening only on some areas (that’s a problem with sharpen ai, you end up with a sharp noiseless area, and noise everywhare else)
Installed the update 1.2.3 from within the program, no issues. First image processed fine, but I selected crop after saving the image and the program crashed. I am mentioning so others can try, since I have not been able to reproduce it.
Currently the subject mask is used to determine foreground and background blurriness, noise, etc. which then influences the Autopilot recommendations. There’s also the toggle in the sharpen filter to only apply to the foreground (the masked subject), in order to preserve bokeh effects or otherwise intentionally OOF backgrounds.
Auto pilot has denoise strength at 1 for this image and still introduces pink into the sky. Setting strength to 10 or turning noise off fixes the problem and produces a good result.
Thanks, Ken. This is a known issue, related to low slider values in the noise removal tool, which we’re looking into. For now, dialing up the Strength slider as you mentioned should work as a workaround for the artifacts to be gone.
Thanks for your clarification, mybe then we need an alpha msk just for sharpening? because typcally I often ony sharpen the eyes, nose edge, mouth / and teh with less opcaity. Therefore I would only select those. But this may mess up the other feature ?
oh, and by the way, for the masks (even the ai one) we need a clear all button. cause often you only want to mask a small part of the subject
also if the sharpening, also in sharpen, could better handle hairs and teeth. Those are the ones that betrays the use of teh sahrpening, they look too sharp (hairs) or weird (teeth)
