Photo Roadmap Update (May 2023)

[eric (Eric Yang)]

Product Updates

Since April’s roadmap update we’ve added a new Sharpen Strong model, added new face selection options, and resolved many long-standing bugs.

New Sharpen Strong model

Topaz Photo AI v1.3 introduced a new Sharpen Strong model that recovers more detail and generates fewer artifacts than before. Autopilot now selects Sharpen Strong instead of the older Motion Blur / Lens Blur options.

Here is a comparison of the results between Sharpen Strong and the old Lens Blur model for a misfocused subject at 200% zoom:

As you can see above, Sharpen Strong helps generate more natural and detailed results for images with severe motion or lens blur. Unlike previous models Sharpen Strong does not handle noise, so you may now need to enable Remove Noise if you notice noisy output.

We will be improving Sharpen Standard and Strong in the coming months. In the meantime, if you still want access to Motion Blur and Lens Blur, you can re-enable them by switching on Preferences → General → Legacy Models. Read the docs for more information and a test sample image.

Allow selecting non-detected faces

You can now manually select faces that weren’t previously detected by Autopilot:

Internally we assign a “confidence score” to the likelihood that each image feature is a face. Autopilot only recovers faces with a high confidence score, otherwise we would end up attempting Face Recovery on many non-face objects.

You can now manually select low-confidence faces to be included by Face Recovery, which should cover many cases that Autopilot previously detected incorrectly. To do this, click the Select button next to the Face item in Autopilot. You can learn more in the docs.

10x faster in-app updater downloads (Win)

We’ve introduced patch updates that make in-app updates significantly faster on Windows. Updating from v1.3.0 → v1.3.x using the in-app updater now requires a 10x smaller installer download than before. A few caveats:

• This only currently affects patch updates (v1.3.1 → v1.3.2). Major (v1 → v2) and minor (v1.3 → v1.4) upgrades will still use the full download.
• For Mac users: we currently only support patch updates on Windows, but Mac installers are already ~70% smaller than their Windows equivalents.

This is a big step towards reducing friction around our weekly product updates. Our long-term goal is to create an update process that you don’t need to think about or work around.

Other improvements and fixes

We’ve also fixed several important issues and made many smaller improvements within the past month:

• Refined the Enhance Resolution Standard model: fix blurriness in highlights, better handle strong compression, more natural faces.
• Fixed half-face recovery issue
• Fixed distorted or rotated masks in Lightroom Classic
• Fixed issue where Autopilot RAW output would be different than preview
• Fixed exporting a large number of images
• Added image thumbnails for all file types
• Improved preview performance by reusing already-processed tiles
• Improved various Autopilot mapping issues
• Added support for HIF extension
• Allowed preserving bit depth on PNG and TIFF output
• Allowed double clicking slider to reset to Autopilot value
• Improved logic for showing the “close image” warning
• Fixed file size estimate issues
• Fixed several crash on exit cases
• Fixed various minor UI issues and bugs
• CLI: allowed access via tpai on Mac, added the -verbose flag, and added better error handling

For a comprehensive changelog, check the release threads for v1.3.3, v1.3.2, v1.3.1, and v1.3.0.

Feature Roadmap

We’re currently focused on improving the user experience around Remove Noise, Sharpen, and several other important workflows:

• Improve the noise reduction experience:
  • Update non-raw Remove Noise to v2, which improves detail retention and slider sensitivity (in beta).
  • Improve RAW Remove Noise by reducing artifacts and enhancing strong noise handling.
• Improve the blur reduction (Sharpen) experience
  • Improve both Standard and Strong model’s blur reduction and detail retention capabilities
  • Resolve ambiguity around using Strong vs Legacy models for best results
• Add a Refine Text filter that mildly restores and removes artifacts from detected text (in beta)
• All models: fix occasional blurry patches caused by tile depth issues
• Reorganize the right panel controls, allow editing detected noise/blur levels, and add Autopilot customization options
• Improve the export flow
• Improve batch processing stability and performance

If you’d like to test new features like Remove Noise and Refine Text, please apply for the beta program. Thanks so much for using Topaz Photo AI, and we look forward to hearing your feedback!

[macz_photo (Markus Zuber)]

I can only say that the idea of introducing the grain option (idea…) would be a great advantage here too. Even with the new model, this bird just looks like plastic.

[henry.richardson]

Does Photo AI (and other Topaz products) use the Apple Silicon Neural Engine when running on Ventura?

I ask because both Adobe and DXO say they cannot use it so use the Apple Silicon GPU until Apple fixes the bug. DXO says it works properly with Monterey, but Ventura is broken.

For example, here is the DXO notice in red.

Eric Chan, one of the longtime Adobe software engineers, wrote this article about the new Lightroom Classic 12.3 Denoise AI function:

Denoise demystified

Denoise Demystified | Adobe Blog

Finally, we built our machine learning models to take full advantage of the latest platform technologies, including NVIDIA’s TensorCores and the Apple Neural Engine. Using these technologies enables our models to run faster on modern hardware.

But, then Ian Lyon wrote:

Re: Denoise AI in 12.3

Currently, the Apple Neural Engine found in M1/M2 silicon Macs is not used by Denoise. It’s expected that performance will be even better when Denoise does use the Apple Neural Engine.

There is no change to above comments regarding the use of Tensor cores in NVidia cards. That is, Tensor cores are used by Denoise.

Apparently, an issue on Apple side means that the Neural Engine is not used by Adobe Denoise. When the issue is addressed, then Adobe Denoise is ready to take advantage of the Neural Engine.

And then Rikk Flohr confirmed it:

Ian Lyons information is correct regarding the Neural Engine on Apple devices

Rikk Flohr - Customer Advocacy: Adobe Photography Products

[ikelsall]

Whilst improvements have been made with Non Raw images, RAW image noise reduction is still an issue. In this image, AI photo is interpreting cloth texture ( see original left) for digital noise. This results in a smearing of detail that should be ignored by the raw tool, because it is not colour or luminance noise.

[ikelsall]

With the same image as a non raw Tiff file, The beta noise reduction tool does not erase the texture detail on the clothing. The Ai engine is activated although this image did not have any notice noise. The RAW noise reduction tool however sees the fabric texture as noise and obliterates the detail. I am confused as to why this happens. Surely the RAW noise reduction should retain better detail than the Non Raw?

[gosa]

Not sure if this is the right thread, but I’m kind of missing seeing how much time each foto takes to process in a queue like in for example Denoise AI.

Would that be to difficult to implement, or am I the only one that things thjat kind of data can be interesting so it’s not even worth the effort?

[Fotomaker]

I’ve mentioned it before (& I realize there’s 'fire fighting" going on to eliminate crashes, etc. that consumes dev time), but I still would like to keep it top of mind as a Wishlist Product Priority to address making the 3rd critical function of PAI (up/down-scaling) operational in the Ps plugin (others can fight for their fave host of choice…). We need the trifecta of products at our disposal in the Ps plugin.

I would regularly use the ‘Gigapixel’ (scaling) component in Ps. But it’s not usable there b/c of the Adobe constraints on “Filters”.

As I’ve mentioned in past requests, I’d be perfectly content either with: (1) the entire PAI operating from Ps’ File > Automate in order to support the resizing functionality and/or (2) splitting the functionality so that the PAI Denoise & Sharpen, etc. functions remain in the Filter > Topaz Labs > PAI hierarchy and there’s a note in that implementation to “go to” File > Automate for the upsizing component of PAI. It honestly wouldn’t bother me to have it bifurcated that way if there is a pointer in the 2/3 element of PAI plugin.

But it is critical to have a smooth, complete product workflow in the Ps plugin vs having to hop in & out of Ps to all sorts of misc. external image processing software.

Thx!

[eric (Eric Yang)]

I hear you on this! Could you go over your typical use case of upscaling/downscaling in Photoshop?

The reason I ask: it’s high engineering complexity to support both a Filter and an Automate plugin, and we found through Gigapixel that most people don’t think to look for a plugin in File → Automate. We could still consider it if it’s a use case that we should support, but I’d also be curious if an alternative implementation could work. For example, we could consider supporting “zooming in” via Filters plugin, where the canvas size stays the same but everything becomes larger.

[Fotomaker]

@eric

I hear you on this! Could you go over your typical use case of upscaling/downscaling in Photoshop?

Hi Eric!

Thx for your note. Yes, I can only imagine how complex it might be to do the s/w design work. It always slays me when ppl say to me, “Can’t you just do thus & so [in Ps]?” The operative word being “just” when they’ve asked for something that involves a lot of skill and steps (at least before the AI announcements…).

But, to answer your question (I hope) … There are 3 ways I’ve used Gigapixel in Ps and that I’d like to be able to use the scaling element of PAI in Ps also. [1] Sometimes I want an older, smaller image of mine to be larger and higher res. So I’ll scale it up to larger pixel dimensions, then immediately process it using my Ps tools (meaning native Ps tools as well as 3rd party plugins). Then print it or share it elsewhere at the upsized pixel dimension (I’d have to do it for various print applications). [2] Sometimes I process my newer, already not bad sized or resolution images using those same types of Ps workspace tools - but then, after the image(s) are processed, do the upscaling/downscaling then. After that scaling is finished - either b4 other Ps processing or after it - I sometimes tweak some more or do some other finishing. [3] I start with either an older smaller image or a more recent larger image (but not medium or large format) - I aggressively crop to isolate one aspect of the image that I want to become an entirely new full-processed, enlarged to a printable scale, image. After I do the cropping (likely using the cropping tool in Ps, but it could be in PAI if that’s my pre-processor) then I’d want to enlarge and do a full range of Ps processing (native & 3rd party tools whether classic Topaz, older Nik, Boris FX, etc.).

@eric

I’d also be curious if an alternative implementation could work. For example, we could consider supporting “zooming in” via Filters plugin, where the canvas size stays the same but everything becomes larger

I’m not sure I’m understanding how you’re meaning “zoom”. So pls bear with me on this. To me, “Zoom” & “Upscale/Downscale” are different things. Often when I’m working in PAI (or any aspect of Ps for that matter), I’ll Zoom in so I can see more magnified what it is that I’m working on to know if I’m achieving the effects I want to. (I usually use CTRL +/- on my Win 11 PC to do that). The scaling I talked about in the just prior paragraph, to me is actually changing the output size of the final image (for ex., I go from a 780 x 1250 to a 3560 x 4250 - even if those aren’t logical relatively, you get my pt - the final output is larger, & capable of being printed larger and with higher res output).

Have I answered your questions wrt what you hoped to find out?

I realize it isn’t always the most intuitive thing for ppl to look for what they consider “filters” in File > Automate. But I think if Ps users are told that’s where the filters appear, that’s where they’ll go to use them. I’ve done that for GAI. I’ve done it for the old Photokit and even the Nik Selective Tool.

I’m wondering how doable (or, klugey) it might be to have a “Topaz Labs” listing in the Ps Filter menu (like where it appears now - or, else done like the betas, with a separate listing). But, then, rather than having a flyout to click on to launch the filter it says, “Go to…File>Automate”. Then if a user forgets that’s where PAI is there’s at least a pointer and they don’t deluge Topaz Spt with (for blankety blank sake, I spent $200 for a filter and it didn’t even load into my Ps…) - not that that would ever happen…