Topaz Photo AI v1.2.7

Hello!

For this week we have quite a few additions and fixes. We’ve added a new preferences panel which gathers up various options from around the app into one place. We’ve also at the same time added some new options, like closing images after they are saved, only selecting faces that are within the currently detected subject, and modifying the default Face Refinement strength.

Along with the new preferences panel we’ve added the ability to decide between 8-bit and 16-bit PNG output, we now save additional metadata when using Lightroom’s Plug-In Extras flow, updated lens correction definitions, and fixed various other bugs. As always more information is in the change log below.


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.


v1.2.7
Released March 16, 2023
Mac: Download
Windows: Download

Changes since 1.2.7:

  • Added preferences panel that replaces scattered settings around the app
    – Includes Autopilot settings and various general settings
  • Added new optional settings in preferences
    – Additional control over which faces are auto-selected
    – Added ability to change default Face Recovery strength
    – Added a toggle for closing images once they are saved
  • Lightroom Plug-In now copies over additional metadata
    – Now copies label, title, caption, gps, and altitude tags
  • Added bit depth options for PNG
    – Preserving file type while saving a PNG will use same bit depth as original
  • Switched to using Huffman compression for PNG
  • Updated lensfun definitions
  • Remove requirement of installing Vulkan on Windows
  • App now cleans up old installers on Mac
  • Changed error prompt when the current version isn’t owned
  • Changed “Cancel Processing” button to be red so it doesn’t visually merge with the “Close Window” button
  • Toggling Face Recovery resets to Autopilot suggested faces instead of selecting all
  • Fixed Wacom tablet not releasing preview after panning
    – Note: You may need to update your Wacom drivers if the issue is still persisting in 1.2.7
  • Fixed crop going out of bounds when XMP crop wasn’t between 0 and 1
  • Fixed temp files staying around when a save error occurs
  • Fixed TensorRT models being slow on certain machines
  • Fixed some Autopilot preferences not applying to already open images
  • Fixed Autopilot settings sometimes not applying correctly due to auto-upscaling
  • Fixed Hasselblad X1D not using the correct camera profile
  • Fixed Autopilot progress going backwards sometimes
  • Fixed View menu zoom options being enabled when no image is open

Known issues:

  • HDR or Monochrome DNGs which previously did not open may have issues when saving as a DNG
    – To get around this you can save as JPEG, PNG, or TIFF which shouldn’t have the same color issues
9 Likes

Great job resolving the issues with the manual brush and movement between zoom , pan and add or remove. Brush feels less laggy and worthy of praise for the code writer. Just FYI different selections are made from plugin to standalone, but easily edited with the brill new brush.
Now for the big hurdle , the RAW colour … looks to me more like the result obtained by a video colourist , and needs a bit of kicking around to get that “photo” look.

3 Likes

Win 11 Pro desktop PC. Processor = Auto (AMD RX6800 XT, Intel i9 12th gen). PAI 1.2.7 (uninstalled 1.2.7b before installing). Standalone version. For this new install & new Prefs setup I decided to leave Faces set to Auto on (I previously had that pref set to OFF).

Hi Adam. Thx to the team for the update. Feedback, potential issue noted below.

Process I Followed: I just installed PAI 1.2.7. After the install launch I opened an image, let the Auto settings do their thing, then I closed the app & re-opened it (I’ve started doing that as a regular habit b/c often none of the settings seem to apply in the Preview pane if I try to run the program after it’s freshly installed - and, I know enough, after all these yrs., to wait for the processing to stop - so that’s not it…).

Context for Believing There’s an Issue: I mention that because - I’ve tried multiple times just now (by closing then re-launching PAI) and keep experiencing the same “issue” with my test image (labelled snips below). Here’s what happens:

What I Observe: When I run the Auto settings (with Face set to Auto Pref), then manually select a Face that A.P. put a blk box around, accept that as Done, then return to the main UI doc space - When I view the image at 100% zoom I see distinct Before & After changes (and the After looks good - nice sharp eyes on the subjects, etc.). But, when I zoom to 200% there is no difference in the quality of the Before/After views. The After at 200% zoom doesn’t look like the After at 100% zoom.

Labelled snips below.
100% Zoom:


I wanted to see the eyes more blown up, so…

200% Zoom:


Per Ken’s (& others prior) comments: The color (and/or tonality) changes in the After processing, in general, even with .tiffs.

Hi,
Sharpening is still the issue too strong even at 1.
Adam message me the new model will be introduced this week? Still waiting .

RAW images …

Sharpen is now good with the standard model.

Masking is still a hit and miss affair where you can see the background included below. And a manual brush to refine without edge-aware is just too difficult as this is supposed to speed up the process.

5 Likes

HI
I tried using standard model instead of lens blur (auto pilot suggests ) it did better job.
Thanks.

1 Like

Have now tried both 1.2.5 and 1.2.7, neither can provide a better result than that in the original NEF file.

What I am seeing after doing as close as the same to the one image with both 1.2.5 and 1.2.7.

Note colour change of blue sky and extra red on the top of the branch.
Distinct reduction of clarity in both trials.

Anyway will upload iamge file and resulting files as well as the comparisons and log files for both attempts at using PAI.

Russell.

1 Like

Hi Don - we are continuously training our subject detection on images where the default mask isn’t accurate. Can you please send me the image to Dropbox? Thank you!

1 Like

Hi Jayesh - was that on the bee image? If it’s another image, it’d be appreciated if you can send it through Dropbox. Thank you!

Files received. Thanks, Russell.

We’re still training the new Sharpen model and it’s currently in beta. I’ve just added you to the beta group. Any feedback on the new Sharpeh model would be very much appreciated.

2 Likes

Hi Yazi, Uploaded filename IMG20210820125505.jpg

1 Like

“Lightroom Plug-In now copies over additional metadata
– Now copies label, title, caption, gps, and altitude tags”

→ GPS is still missing for me (Lightroom Classic 12.2.1, Nikon Z9 .NEF)

Using the Photoshop plugin, Sharpening results in a noticeable pixel shift.

1 Like

RAW to DNG workflow is still unusable with multiple bugs…

  1. If 3 or more masks are created, only 2 are transferred to the DNG.

  2. When I crop in Lightroom, it is not transferring to the DNG.

  3. Unwanted color artificers are introduced in DNG.

  4. Whiskers (fine hair) pixelated (remove noise at Normal and 1 for both Strength and Detail).

  5. The Mask has the wrong orientation on some vertical images.

The current workaround is to use TIFF/JPG instead of RAW workflow.

I’ve uploaded two images (with XMP files) to the Dropbox folder. Can someone please advise if these bug fixes will be fixed in the near future (weeks or months)?

Further to my earlier post about colour, there is a very noticeable change from the AI working screen to the tiff saved and opened in PS. Perhaps this screen grab of the two images, one in PAI and the tiff generated in PS 6


I think it is important for the working screen to match the output .

Hi Yazi,
Same Bumble Bee image, I have seen improvement ,now it will only sharpen what I want not whole image , I think lens blur, and motion blur sharpen is too strong even at 1 on this particular image whereas Standard Model is working as intended .
Bee image is perfect example to try it out.Look forward to try out New sharpen model .

Doesn’t PAI auto upscale low res images to 12mp? This never happens for me. I dropped a note about in the last beta. Here is a screen shot of the upscale panel in 1.2.7 Release version

scale

It is already answered in the Beta, don’t post in multiple localions.