Topaz Photo AI v3.3.2 - v3.3.3

[Lingyu]

Hello!

We are releasing the v3.3.3 hotfix on Mac only to resolve the Apple Photos plugin crash.

Another patch today, with the following changes:

• new authentication flow for trial users
• add support for Canon R5 Mark II and Nikon Z6 III RAW files
• allow saving with crop and no filters enabled
• various bug fixes

You can find the installer links and full changelog below. For any issues please reach out to help@topazlabs.com for support.

---

v3.3.3 (Mac only)
Released November 18th, 2024
Mac: Download

Changelog

• Fix Apple Photos plugin crash caused by deeplink feature

---

v3.3.2
Released November 15th, 2024
Windows: Download
Mac: Download

Changelog

• Require authentication for trials
• Add support for new cameras, including Canon R5 Mark II and Nikon Z6 III
• Show update dialog before trial mode button
• Allow saving with crop and no filters enabled
• Add deeplinks for Windows and Mac
• Fix text recovery skipping over controls tab hitting esc
• Fix drag and drop folder on app executable
• Remove denoise deactivation message in upscale
• Fix reset to default bug in Autopilot preferences
• Fix applying current settings allowing RAW filter to get doubled up
• Fix applying current settings to linear DNGs adding RAW
• Fix upscale not working for some images in batch
• Update Exiftool
• Automatic Lensfun Update

Known issues

• High efficiency (HE*) compression Nikon Z8 and Z9 files have color shift
• Crash when using Photo AI as a plugin from Lightroom Classic on some Windows 11 devices
• Crash on some macOS 15 (Sequoia) devices with face detection on. Please disable face detection in Preferences > Recover Faces to prevent the crash.

---

Lingyu Kong
Technical Product Manager
Image AI

[Fotomaker]

Win 11 Pro (desktop PC). Processsor = AMD RX6800 XT. Standalone & Ps (2024) Plugin - both via File > Automate & Filter Menu.

Superficial run through (usability check) … with .jpg image I don’t consider large (i.e., 900 x 1256 px).

JPG image test run = composite comprised of: base “straight” photo, Firefly gen AI element (tree), stock image content moved onto background of straight photo. All of the components that were composited in were low res.

PAI Standalone - Launched fine. Able to process image using SuperFocus only. I manually selected the objects in the scene and not the background wall. SuperFocus added definition to evergreen needles that were a mush (tech term!) from Firefly gen AI.
Took ~ 5 mins. to process on my PC.

PAI Plugin (via Ps File > Automate): Launched fine. Able to process image using “normal” AI sharpening models - my manual choices and settings, not AP. Didn’t do as good a job on the faux Firefly gen AI tree needles as SuperFocus had. Essentially crisped up the pixelated mush a bit. Wouldn’t be usable with those sharpening models only for this type of photographic composite. Able to save back to layer stack with no issues.

PAI Plugin (via Ps Filter Menu): Launched fine. Essentially replicated same workflow as prior plugin just to be sure usable via either access method. Same results (except Filter menu overwrites Ps launch layer - for that reason I prefer using the Automate approach - plus can scale via File > Automate).

Snip of image tested:

Bottom Line: SuperFocus was necessary to make low res gen AI (Adobe Firefly) usable as part of this test composite. Firefly isn’t the best calibre relative to the various gen AI options out there. Something like this feature is needed as ‘salvage’ for the low quality AI images Firefly produces.

Now that image has 1st stage processing completed and is (handily) in my Ps layer stack I can make additional Ps adjustments… PAI has no negative impact on other Ps adjustments:

[plugsnpixels]

Playing with Refocus on the M2 Mac Mini, it’s decently fast! Using film scans as sources.

1974 Kodak Instamatic (126 film) image that I developed myself at the time:

Colorized in PS (I wouldn’t complain if PAI could do this as well! Trade it for Balance Color…).

Various others from 1990s 35mm film with mixed results but all with improvements over the originals:

[mike162]

When are you going to fix the Lightroom Classic bug … and it happens with local only files despite what you guys say? It’s ridiculous that with that and colour casts I’ve not been able to upgrade for now 5 releases

[thomas.acito]

my main issue with photo ai is that I cannot update from within the app. Download is fine but error appears when I select update.

[plugsnpixels]

REMOVE: Photo AI vs. The Competition!

I’ve got the latest releases from the various competitors and here are the results from trying to remove the tree!

COMPETITOR #1

Already in trouble…

Tried again, not worth the trouble…

COMPETITOR #2

Technically, the tree is gone-!

COMPETITOR #3 (Photoshop, first 3 results):

PHOTO AI, first try:

So it seems PAI is second only to Photoshop for this particular task.

In other news, you need to leave PAI to do any additional artsy, which Topaz Studio used to do…

[oneh]

no option to set
“FIX Compression” to zero or set it to any value
so if i load 20 images these option always different and i must set all manualy to my value

[Paul_Coddington]

Imagine an AI colorization feature that intelligently identified objects and allowed you to assign colors (shirt=blue) rather than merely estimate them from training data, etc.

That would be brilliant if someone pulled that off.

[plugsnpixels]

Even simple text prompts with the image open should be able to do this eventually, as with image-to-image rendering that we already do.

[magnus.nilsson]

Yeap the Lightroom crash started on v3.2.1 in. It was just a few weeks after I bought Photo AI , and the crash continues with v3.2.2, v3.3.0, v3.3.1 and v3.3.2…
So staying on v3.2.0.

[axel.matthies]

I’ve seen some color casts on using PAI 3.2.0 with Lightroom. This was fixed since version 3.2.1.
I also don’t see any crashes on using PAI with Lightroom.

[jurgen.wietschke (juergen.wietschke)]

Ich habe das Problem, wenn ich 2 mal sharpen hintereinander anwende, dann stürzt PAI ab. Das Gleiche passiert auf verschiedenen Computer. Kann ganz einfach reproduziert werden.
So ist das Programm nur eingeschränkt nutzbar.
Als Betriebssystem läuft auf beiden Computern Windows 10 64 bit.
Ich habe versuchsweise mal PAI 2.4.2 installiert. Dort kommt es nicht zum Absturz.

[Artisan-West]

Using your photo, I ran a quick test to remove the tree with PAI and one Competitor (that I own). See below:

PAI:

competitor:

[plugsnpixels]

The Lost Temple! Interesting you got a replacement instead of a removal, though you are using my artsy version FWIW.

[Fotomaker]

@dakota.wixom @yuan.liang

If we have what’s tagged a “Large Image” that we want to SuperFocus (& it’s taking a long time even to generate a render ETA - if it does at all), can we realistically create a series of crops of the full image, SuperFocus them individually, then ‘glue’ them back together to form a super focused version of the original image?

Does the SuperFocus model process consistently enough to be able to reassemble individually SuperFocused segments into a coherent semblance of the original image - but much sharper?

Or, does it make up the appearance of elements and would do so relative to the individual crops in ways that wouldn’t look like the original image if reassembled?

Does that question make sense?

Needless to say, if the answer is “yes” that would work - then I’ll want to know if I can create an action to work through those steps for me.

I’m trying to deal with (innovate) using SuperFocus on a Win 11 system that takes forever to process with it (so, realistically I can’t use it for typical images), but I’d like to be able to use the feature on images other than teeny ones. Thx.

[plugsnpixels]

This is an interesting question!

On my M2 Mac Mini I am setting out to do this right now with a film scan (4134x2638px @ 2900ppi). Maybe you recognize this place?

In PS I divided the image into 9 rough sections and saved out each one into separate files.

In PAI I was hoping to batch process these but it wasn’t clear that the same setting (I did not choose anything except Refocus, maybe Autopilot is doing its thing) would be applied to each, maybe it would have, but I cancelled batch and went manual:

It’s taking about 2-3 minutes per render. As each finishes, I see they are exported at different magnifications (2.9, 3.3, 3.4, etc.) which will make it interesting to reassemble. I did not see an option to control the output size – but I just noticed halfway through that I could turn off Upscale… However Upscaling is an expected and automatic part of the Refocus feature, I believe.

I tried a second time to batch process while purposely turning off anything I could except Refocus and making sure Autopilot was not checked. The list of upcoming treatments were not consistent.

The renders banged out in seconds and had the same dimensions and file sizes as the individual renders from the first try (except the last one-?), and the treatments were not consistent, as feared:

So that didn’t work at all and this set was deleted.

I know one can batch process images successfully as I did this A LOT with Gigapixel’s Redefine, but this is Photo AI.

Now let’s try to reassemble the first results in PS.

Due to the different output sizes, a manual reconstruction was not viable:

How about Photomerge? This failed before I remembered that you need a decent amount of overlap:

All told, it’s probably quicker to just Refocus the entire image-! I am doing this now and we’ll see what level of pain is involved… (looks like 15 minutes, maybe even less than the separate renders in total).

The unretouched original:

So at least on an M2 Mac Mini with an image this size, Refocus is survivable ;-). M4 Mini’s are out now, great deals!

[Fotomaker]

It ended up looking like a Hockney collage after being reassembled. Hmmm. So, guess that may not work due to the output sizing.

I also tried Exporting an image at more than 1/4 its size from Ps. Then bringing that back in to Ps PAI plugin to try to SuperFocus it (it ran for 2 mins. at that size) and instead it made it look Impressionist vs sharper. So maybe I did something wrong. Or, maybe the starting image was too sharp. I decided to try unsharp mask on it (in Ps) instead… this is how that turned out (a shot of mine from eastern Sierras - where the fall colors are mostly yellows, vs the yellows, oranges, peaches, reds I grew up with on the east coast). This is the downsized jpg - it tends to give it a darker look and almost more illustration-y in the conversion:

[plugsnpixels]

Both Refocus and Redefine are using AI to reinterpret the input image, not just “focus” or clarify, etc. That’s where the artsy is coming from. I don’t see a way to avoid it or minimize it down close to zero as there is little control at this point. So I just run with it

[jason-2269 (JasonNewton)]

Deeplinks are mentioned in the release notes. What exactly are they?

[axel.matthies]

Das Problem habe ich auch. Ist mir bei der Version 3.3.1 (unter Windows 11 23H2) das erste Mal aufgefallen. Ich nutze diese Konstellation nicht so oft, deswegen stört es mich nicht so groß bei meiner Arbeit. Allerdings sollte der Fehler schon zeitnah behoben werden.
Die Version 3.3.2 habe ich noch nicht getestet.

I have the same problem. I first noticed it with version 3.3.1 (under Windows 11 23H2). I don’t use this configuration very often, so it doesn’t really bother me in my work. However, the error should be fixed soon.
I haven’t tested version 3.3.2 yet.