Topaz Photo AI v0.4

Hello everyone!

This week’s early access update features some updated auto pilot logic as well as various fixes across the app. We’re currently working on refining the auto pilot logic and design even more, so expect some more changes next week as well!

If you own DeNoise, Sharpen, and Gigapixel with up-to-date licenses then you have access to Photo AI. You will need to claim a Photo AI license by going to your account before you’ll be able to login.

Released July 28nd, 2022
Mac: Download
Windows: Download

This is an early access release, so there will be some issues you may run across. Issue reports and general feedback are much appreciated. If there’s a specific image you’d like us to see, you can send it to us at this dropbox link. Below the changelog are a list of issues we’re already aware of.

Changes since 0.3.0:

  • Tweaked autopilot logic
  • Show original file size in resize panel and export list
  • Added TensortRT model for RAW DeNoise
  • Added total face count to the autopilot list
  • Added download speeds in installer for Windows and Mac
  • Last view mode should now be saved between runs
  • Only show autopilot image requests when autopilot has been changed on an image


  • Fix white output and some line artifact issues on Windows
  • Fix autopilot not detecting faces
  • Fix autopilot saying all images are not noisy
  • Fix some images appearing rotated 90deg in thumbnails
  • Fix not opening images by dragging if they’d been previously opened and closed in that instance
  • Fix autopilot not always re-calculating when switching images
  • Fix RAW images failing to save in external editor mode (will now save as TIFFs)
  • Fix scale being reset to 1 when switching images
  • Fix switching images sometimes not changing the blue highlight in the file list
  • Fix scale being off until the first preview is processed after switching images
  • Fix certain RAW images opening with wrong orientation

Known Issues:

  • Use these Settings for All not applying to all images sometimes (tends to happen more when using this option on an image that isn’t the first)
  • Exporting a lot of files may not go through the entire list (stops early, stuck saving, etc)
  • Batch film strip may behave oddly in some cases (resetting scroll, not highlighting selected image, etc)
  • Some RAW images with specific orientations may not have the correct orientation on open
    – Specifically, images with orientations that are just vertical/horizontal mirrors may remain mirrored on open. 90/270 degree turns should open fine.
  • Auto pilot may not detect every face in an image

Win 11 Pro. Desktop PC. AMD processor - setting in PAI.

Testing Standalone.

Perceptually, it felt like this rel. downloaded/installed faster than previous releases. Installed with no issue over .3 rel. (I didn’t uninstall .3). :+1:t2:

Tried a Drag & Drop jpg into the UI. So I can try it again a few times per Changelog fix list… I have only done once so far. Not repeated to ensure repeatable several times. Also haven’t closed PAI to see if View type is maintained.

Used Autopilot for 1st run - It produced a new msg that it detected “medium” noise level in test image. Had never gotten that assessment before from the Pilot - previously only “low” noise detected.

However, I still got an “Image is in Focus” assessment. Which I don’t agree with for the test photo (see labelled snips below). I turned Autopilot off and made a Manual Sharpening slider adjustment to where the image did appear in focus (with the manual sharpen slider, it was a very nice sharpening result! So if that could happen in Autopilot it would be great. But, as someone who likes some processing control, it’s good to know I can get desired results with the sliders (and control where they’re applying sharpening - ultimately - when masking is added…).

For Sharpening - I tried both the slider pull and typing in a value - both worked okay.

For the final step (again, snips below) I added enhance - natural to the the settings that PAI gen’d and what I set in Sharpen. It may have added some incremental clarity. As a bonus. It wasn’t a radical change - which I think for this photo was perfectly fine.

BTW, I’m not seeing any serious color shifts from top to bottom (untouched to manipulated) in the snipped previews stack attached. There may be some brightening or desat on the pavement in the 2 snips in which sharpening was manually added. See what you think.

Quick note. Closed PAI .4 after test above.

Re-opened. It held the Split View I was working with.

Was able to drag & drop same cab image in a 2nd time with no issue yet.

Also - Able to launch PAI .4 okay from Ps 2022 as Plugin. And, it held the Split View I worked with using the Standalone into the Plugin too. :+1:t2:

Download was much quicker with 4 and made me smile :smiley:
Opened a CR2 into standalone with Auto settings , accepted and processed. First image posted.
Then I opened the same file in PS6 and developed in ACR with no sharpening .

First noticed the underexposure in AI 4 was only around 1/2 stop , the original in PS required a RAW setting of 1/3 under … so AI is still darkening. Tiff file from AI was again not recognising lens data in PS , needed 1/2 stop+ exposure and lots more contrast and black crushing than the original . How come the preserve format command does not create a DNG ?

Leave you to decide on the sharpening in AI 4 , to me without masking it is dangerous , it even attacked completely oof areas , see the new bokeh in the sky area to top.

Last but not least the output from AI4 opened in ACR with all zero settings to show how it looks as is.

All images sent in full size jpeg to dropbox , as well as an infocus clip of the white car so you can see what in its shape threw the sharpen.

Win 11 - Ps 2022 PAI .4 Plugin. Split View (retained from previous tests)

Decided to stress test Autopilot face ID & Face Recovery.

It gave me a count of 38 faces detected in the group photo. I’m gonna have to trust it b/c it has more patience than I do to count them all!!!

I don’t know how what kind of coding might be required, but it could be very cool to have an (optional) feature like Ps added in one of the recent updates for Selections — in which we can choose (or, not) to have an overlay put over the Subjects (in this case faces) it detects in a scene - and then decide whether to have the processing applied globally to all those selected subjects/faces or selectively pick the ones we do want the processing applied to.

Get an inch, want a mile syndrome …

p.s. I just sent this pic via Dropbox in case you are interested to figure out which faces autopilot detected. :slight_smile: I wonder if it got the guy lying down sideways in the front but maybe didn’t get the misc disembodied heads behind the 1st couple rows of ppl.

I’ve been running Gigapixel AI from Lightroom Classic to denoise, sharpen, and enhance the faces in old family B&W scanned photos. The face recovery in particular is impressive, but it takes a long time to render the different options. I was hoping face recovery in particular would be quicker with Photo AI.

Overall, my results are mixed.

When Photo AI determines the faces are “low quality” it quickly does an excellent job of face recovery while also removing noise. Wow! I was amazed at the speed and the quality of the results.

Unfortunately, Photo AI thinks the faces in most of my old photos are not low enough quality and they are left as is. Please add a way to adjust what “quality” triggers face recovery.

Are the auto-detected filters accurate? Yes, but I wish the “auto-detect” was adjustable.

For some reason the first try hangs up. Happened for several versions now. Windows 11. Second try seems fine. Output from RAW file is still sRGB.

Autopilot is a bit patchy. Some areas sharpened OK but not others. File uploaded.

Still when changing the display mode, i.e. split screen to full screen etc., the image is recentered rather than applying the

Split Screen:

Switch to full screen mode:

Finally got around to downloading and trying… Download went quickly…install took awhile but understandable as I watched all the things being added. Windows 11, new computer, I immediately dragged several photos in and the processing was very quick. It improved nearly every photo I tried…some were bad to begin with and some were quite good, but in all cases I could see improvement…not unlike AI CLEAR, which I love. I did find one interesting set of photos, however and I don’t know the reason for different processing on essentially two consecutive photos. look at the jacket collar above her left shoulder. The shot where she looks away is awesome…the glasses cleaned up, the face, too. but the one where she is looking right at me has a blobby/blurry collar. found that to be odd. Perhaps it’s because on the one where she’s looking at me, it’s a bit darker and not as crisp…but the program really attacked it… and the headphones actually got worse on the straight on one.

Exciting new product! I am doing the artwork for my new album and even though it’s illustration not photos the first tries are good!

Intel MacBook Pro (Monterey): Download, install and subsequent downloads went well after manually deleting previous beta. Processing is painful on Intel (2013 i7) :wink:

M1 MacBook Pro (Ventura beta 3): Download and install went well, it didn’t perform subsequent downloads (I forgot to delete previous beta FWIW, but it’s gone from Applications folder now).

M1: After turning off Autopilot I wanted to see what Enhance did on its own, so I turned off Remove Noise. Noise removal did not stop, I had to wait it out.

With only a 38 meg JPEG, Photo AI seems oddly slow for some reason. I don’t know if this has anything to do with the post-install downloads not taking place. The previous beta zipped through tasks in seconds.

I’m going to run the pending Ventura b4 install and try again.

I’m noticing PAI now defaults to Autopilot Off after closing an image and opening the same or another image, when Autopilot was just On a moment before. We had some discussion about this earlier, was there a change?

The noise in my iPhone photo is not “severe” – you wanna see severe, I can show you that, ha!

What,s error code 2755

You need to provide details of your OS, RAM, GPU, dedicated vRAM and where the message came from and what you were doing at the time.

If it is Windows it usually means Windows cannot open the files or lacks the proper permissions. For example if an installer fails it may mean you need admin permission to do the install … i.e. run it as an administrator.

Quick tests with 0.4
looks going forward :smiley:
Just wanted to report :
It’s annoying when we close the save windows the preview screen is not updated until we change setting or move the display ?
When we load several files it’s looks going better but still some issues . When I load a second file process it with autopilot and close the first file the display hang up with the message No filters active whatever we are doing on the control panel . Eventually it came back to work after I did a save which didn’t process any of the filters selected
With this picture only two faces on 3 were detected by Autopilot , probably the profile face was missed ?
in another one 6 front faces was correctly detected good work

Still something I spoke with you
Not very serious issue but perhaps symptomatic
When I use the PrintScreen key on my Keyboard under Windows 11 I got strange results
none of my photos software are doing that

With this RAW file
[images](DSCN3259_DxOPL5-DP20-on1nn.tif - Google Drive, DSCN3259.NRW - Google Drive)

I don’t agree with the Autopilot
with the RAW file it tells medium noise Dn11 sh 55
and with the TIF denoised (dxo and on1) it tells severe noise Dn2 sh 23 so why severe ??

the results are now on par with Gigapixel, even tough the AI sometimes misses parts of the hair to enhance. I just wanted to note that you made three of my family very happy when I made large high quality photo prints of some upscaled black and white photos of their childhood. :blush:

Im trying out the new photo AI.
It starts to fix noise but when its about to start sharpening it stops, nothing happends and it says, down in the left corner, “no active filters” ?

Looks like its not dealing with black and white images of faces.

Also not shown in the photo I still get a lot of sharpening halo’s especially around text.