Photo Roadmap Update (Jan 2023)

Since the last roadmap update, we’ve delivered 3 roadmap items in 5 releases: improved blur estimation, strong raw noise parameter estimation, and improved batch processing workflow with Apply Current Settings. We’ve also shipped some significant performance and stability improvements. You can learn more about these changes here.


We’re actively working on model quality and workflow improvements:

  • Enhance Resolution model that recovers finer detail is currently in the final stages. We will deliver it in Gigapixel first, then Topaz Photo AI shortly afterwards.
  • Remove Noise model for non-raw images that recovers more detail and responds better to slider changes.
  • Sharpen model that’s intended to produce fewer artifacts.
  • In-app cropping tool.
  • Apple Photos extension.

Next up:

  • Improve initial subject detection model accuracy, particularly the Portrait option.
  • Improve raw file handling, particularly around lens correction, metadata, orientation, color, and interaction with Lightroom. (Please reply and upload your raw file if you’re having related problems.)
  • Explore fixing the “pop-up background face” issue where Face Recovery will create jarringly sharp faces from a blurry background
  • Further improve subject refinement capabilities, especially for images that are difficult to mask with AI Brush [1] [2] [3]
  • Updating Autopilot Configuration affects open images instead of forcing an app restart [1]

Note that we canceled adding an Autopilot Configuration option for noise until we can better assess the quality of the upcoming Remove Noise model.

The future?

Topaz Photo AI is designed to maximize image quality on autopilot. Our vision is to build a tool that you trust to produce outstanding image quality with zero clicks, so you can focus on more creative parts of your work. To help guide our future roadmap for this, please give us your thoughts on:

  1. Image quality. After the current round of upscaling, denoising, and sharpening improvements, which image quality problems do you wish we would solve?
  2. Image understanding. How does Autopilot currently misunderstand your photo or give faulty recommendations?
  3. Workflow. Where does Topaz Photo AI currently cause the most pain in your workflow?

In general, we focus on building features that pave the path to zero-click image quality for as many users as possible. (This is also why stability and performance are so important to us.) While we won’t be able to implement every feature request, we do read each post and consider your feedback invaluable.

Thanks for using Topaz Photo AI, and we’re looking forward to building a great tool in 2023 together!


Stil no DUMB brush option ?
We absolutely need the dumb brush option, I see no good at all in the AI brush, its both awfull to use and useless


Eric, It sounds as if there are absolutely no plans (or, intention?) to add the Masking Brush features so many customers are requesting, and that I prepared a detailed Wishlist regarding. Last I looked on the prior Roadmap thread, at least 19 people had liked/agreed with that Masking Brush features wishlist (& others reinforced it by agreeing in their own posts)…

Is the bottom line that there will not be improved masking brush & manual selection features in PAI this 1-2Q of 2023?

Hi @vincent.lugnier and @Fotomaker , we hear you loud and clear! In the original post there’s this line addressing the feedback about the AI Brush:


Yay! Thanks Anthony!!! Glad to know… :clap:t3: :pray:t3:

1 Like

Thanks Anthony, I saw this line indeed, but wanted we make sure we won’t just get a smarter AI brush instead of the humble and needed dumb one.
I’m reassured, thanks

1 Like

When I bring in a photo as a stand-alone, I get the option to add a Photo AI prefix to the filename. I don’t see that option when going as a plug-in from Lightroom. I would like to be able to add the prefix there also before processing starts. I sometime forget which images have been run through Photo AI and which haven’t. Prefix option in filename would be a big help.


So far the main problem I encountered is that the Noise removal thinks textures (on the wall, the table, on the ground …) are noises. That’s quite disturbing, especially if one intend to do a batch processing.

  • Upscaling seems to treat certain vegetation (grasses/bushes) and dirt/gravel paths as if it is fur.
    (the amount of detail it can recover/fill in and the blur reduction is impressive btw!)
  • Denoise treats wall textures (e.g. stucco or marble) and clothing fabric patterns as if it were noise and removes them.
  • Denoise strength is on the high side, glossing over details and creating big smooth uniform patches. The strength slider has 0 effect on the strength. So currently there’s no way to prevent this artificial ‘painting’ look.
  • As said above, the default denoise strength in my opinion is a bit too high. I would probably no longer be bothered by the default strength after you implement the announced improvement "Remove Noise model for non-raw images that recovers more detail and responds better to slider changes.".
  • Can’t really blame it, but: it recognizes faces on statues and tries to recover them as if it were a human, giving them hair textures and (brown) eyes.
  • pain is too strong a description, but it would be nice if the plugin roundtrip (from Capture One in my case) could also append postfix labels to the saved (tiff) file name like the standalone app does.

Just for the record, I really like the AI brush. It has been a huge time saver when trying to clean up some of my wife’s scuba diving photos. I’ve also found it pretty good with my wildlife photos.
Yes it absolutely needs a ‘dumb’ brush in addition. There are some things the AI brush can’t interpret. And sometime a dumb brush is just better and easier.
It would be nice to have an option to run both the subject detection and the AI Brush after applying the denoise, as I think this would make the object detection better. Same applies to any new dumb brush.
It would also be nice if there was an option to fill an enclosed mask, so I mask around the edges, and it fills in the middle.
But please do not abandon the AI brush just because a few people don’t like it. We don’t all think it is “awful” and “useless”.

That is more than welcome, especially with the loss of detail at the moment.

Unfortunately RAW file handling is a real problem … but without intervention where we can specify a tone curve and other basic adjustments I don’t see it improving. AI is all well and good but RAW conversion is a personal like/dislike and that is why the converter we use is a converter of choice.

I think what we are really looking for is an initial detection by AI and then manual adjustment by a brush with controls for fill, edge-aware, hardness, opacity, flow and (if possible) and/or air brush.

To put it simply the current RAW conversions and/or Autopilot cannot interpret individual likes and dislikes such as the ‘mood’ of an image. And the most pain in workflow is the inability to do manual adjustments to ‘tweak’ the results.


Once again, I urge putting improvements to the masking brush higher on the agenda. As things stand, SAI does a better job when refining subject selection than does PAI, even though the initial AI algorithm seems to be the same.
Masking in PAI does have the potential of being excellent. Therefore (and I have said this before) it would be great if the masks could be saved, not only to enable one to return to them but also so they can be exported and used in other programmes.

1 Like

Please address shifts in color and/or exposure variation depending on save format selected.


Yes, but also ability to remove areas that may have been auto-detected. Or, to have the flexibility of an On/Off switch to activate auto-detect if we only want to manually, selectively paint a small area for PAI to work on.

And, a Masked Area invert function too, pls.


I would really like to see constant improvements to face recovery. I’ve tested Photo AI vs GFPGan and GFPGan almost always produces more natural and accurate faces. GFPGan typically can fix a face so that it is clear, but also free of any weird pixels or things that are obviously not part of a real face. For example, sometimes the eye pupils and color in Photo AI upscaled images have a weird blur on them or splotch, or they are not a round shape. These are things that are not part of a real human face, and yet they are sometimes included in the upscale.

GFPGan is able to fix these issues because it is better at knowing what an eye should look like. Photo AI is real close to the same quality, but in my opinion it’s not quite there yet.

If GFPGan is an 8/10 with face upscale accuracy and improvement, Photo AI is a 6.5 or 7. It’s very close. I’d love to see this constantly improved so that Photo AI is the best face reconstruction tool out there.


Thank you for the detailed update!

  1. Text Recovery: A way for Photo AI to recognize small and/or blurry text and upscale the letters. Like the current Face Recovery feature, it could be turned on and off for individual images and text regions as desired. The text might need to be editable so we can correct any misread text. Also, it may only work for one or a few languages and not all font styles at first.
  2. The Face Recovery feature is great for small faces while the usual upscaling becomes more effective for medium and large faces. Currently there is no easy way to have Photo AI use Face Recovery on the “small enough” faces while the larger ones are left alone. I’d like to see an Autopilot option to automatically apply Face Recovery only to face sizes below a certain threshold (of course individual face selections can be manually overridden).
  3. The two main “pain points” for me are batch processing and extra clicks in the UI.

Batch processing: Unfortunately, the queue in Photo AI (and Video AI) leaves much to be desired. For my usage, it has generally gotten worse over the years with the first version of Gigapixel which I ever used (4.3? 4.4?) still being the best among the Topaz apps that I’ve used. Also, the Autopilot isn’t good enough for me to just “set it and forget it”—I still have to make some manual checks or adjustments. So I typically end up preferring Gigapixel AI over Photo AI if I want to enhance a double digit number (or more) of images at once. Basically,

Gigapixel AI 4.4’s queue > Gigapixel AI 6.2’s queue > Photo AI’s queue.

In Photo AI’s queue, basic interactions such as multiple select, drag to select, selection checkmarks, and reordering are not present, and one cannot see the detailed settings for many images at once. The icon view in the queue is useful but only one full row is visible at a time, making it inconvenient when working with a large number of images. In my opinion, it is better to make the queue work like a file browser, because not only are file browsers designed to work with many files at once, users already have experience with them. If Photo AI adds file browser-like functionality in the queue (Handbrake’s queue is a good example), then users can select, reorder, and apply settings using similar click and drag operations as they are familiar with.

Even though Gigapixel AI 4.4’s queue was nowhere near perfect, it at least filled up half the window (excluding the sidebar) so it wasn’t cramped, and I could add new images to the queue even while the app was running.

The upscale dialog only has 1x/2x/4x/Max and text boxes for resolution. I’d like to see additional options which are convenient in some situations.

  • Set the longest side or shortest side to a certain number of pixels.
  • Type in a decimal number to upscale by this factor.
  • Enter a fraction to upscale by this factor (useful if the desired upscale ratio is complicated or an infinite decimal).
  • Affordances for chaining models. In my experience, upscaling a poor quality image after applying sharpen/denoise through Topaz may result in a lot of artifacts. I typically upscale images 4x in most situations, but I’d like an option where images that are only upscaled (ignoring Face Recovery) are upscaled 4x while images that go through multiple models (e.g. sharpen + upscale) are upscaled only 2x at the end.

Extra clicks: The drop-down menus for “Recover Faces” and “Enhance Resolution” should default to visible, not hidden. Displays, even many laptop displays, have high resolutions, so there’s little reason to avoid displaying more information directly on the interface, even if it takes up some space. Also, a slider and individual buttons for the zoom percentages would be convenient.


Also, regarding question 3 about workflow. The biggest pain point right now is speed. Each image just takes too long to upscale and save. Since the video product can’t fix faces yet, I sometimes try to export the video as frames and then upscale using photo ai. I know this isn’t perfect but it’s better than nothing. It just simply takes too long with thousands of frames and the program in the past could not handle thousands of photos being batch processed without crashing.

Just some feedback on my new user experiences.

I recently stumbled across gigapixel AI and after using it for an hour or so I purchased Topaz photo AI accidentally intending to purchase the bundle and not realizing it was a different program. I had run into quite a few images where the enhancement really was not fantastic and needed further refinement, but I hadn’t request a product exchange cause I do like the idea of the 3 programs in 1. I contacted support and found that if you purchase the photo bundle you can get photo AI, upon requesting licenses for the bundle I was given them. 5 stars for customer service on that end.

That leads to a big however of until Photo AI is truly a no button option there are tons of features that need to be added if you don’t want other people rightfully complaining. The enhancement on photo AI has no sort of slider or any options for refinement at all. When denoising and upscaling an image the results can be unsatisfying and just upscaling low res images can turn into Lovecraftian horror shows. I was able to get good results with Gigapixel after some refinement, and if I wasn’t just given gigapixel from my purchase I would have requested a refund of photo AI.

I know the AI is being refined but when results aren’t up to par there needs to be options for refinement, not just “go buy our other software”, more user refinement options is a good thing while the AI advances. Information from someone tweaking settings can be further used to help refine your model too. in the end realize ZERO CLICKS SHOULDN’T MEAN ZERO CONTROL

Additionally across the board all of the applications with face enhancement need additional refinement and options. I can upscale an image in photo AI and while the subjects hair comes out fantastic, when I turn on face AI the hair in areas effected by the face correction now look bad, so I have to export the file twice to photoshop their face back. Being able to mask face enhancement would allow users to get better results with possibly integration with the model trained from Mask AI.

Also the post # 37/55 from the Dec. 2022 Roadmap, pls. That’s the post w/19 :hearts:

The other ref numbers listed above just seem to say a “dumb brush” is needed. Rather than a manual brush with specific manipulation features.