Refocus on core features (Gigapixel/DeNoise/Sharpen) - No more fancy features

No this is another „Grid“-Patches Bug, but has nothing to do with the „Blurry“-Patches Bug :wink:

1 Like

It’s quite impressive to see how far the technology and especially concrete applications have come…
I remember experimenting in Matlab with FuzzyLogic and Artificial neural networks for some abstract university tasks… :crazy_face:

1 Like

I want more features not less. I want more power to edit color and lighting. One of the reasons I don’t use Topaz for RAW development is I don’t like the color and lighting rendition.

Hello @tom.cass , welcome!
This thread is not meant in the way - no more features - ever :wink: All we are trying to convey is - we need a stable and well functioning product first before it is stuffed full of even more features.

We all bought a subscription at one point and feel we were left hanging. The old well working Gigapixel/Sharpen/DeNoise products have been discontinued in terms of development and we are left with Photo AI which is giving us a lot of problems and deficiencies.

Every new version each week has yet other issues - while Topaz Labs keeps pushing for more and more features. That’s all we are trying to say here. First consolidation and quality control and then more features :slight_smile:

For us it’s a matter of either renewing our subscription or not…


I have been using Photo AI for about a year. I have never experienced a single one of the bug problems that are addressed in the regular upgrades. The stand-alone products, which I no longer use, have been very problematic and often unstable.

1 Like

It’s good to meet a satisfied customer, too :smiley:

My experiences were different. Many of the weekly new versions were not usable for me - in four releases alone the CLI wouldn’t work at all, and then there was always some single issue that got me going back to a previously well good known version, like 1.3.2 or 1.5.4. And with the latest there is even an update issue and double face detection issue.

It often feels like we are presented with beta versions (that we have to sort out, if they are usable or not) - not tested release versions. Often enough there are also hotfix releases a day after the actual release to fix some even newer bug or problem, that was unforeseen.

There are technical processes in the field of software engineering that can be established to ensure quality control: with continued development of unit, integration and functional tests. But it takes time to build and maintain such a test base (taking time away from features). But once in place that would test most of the functionality on an automated level, so that every future release can be considered a working release.

Then there’s the matter, that I work with a lot of photos at a time (hundreds and even thousands), it’s no fun working with that amount in Photo AI (UI, functions & batch are not cut out for that).
And then there are long known core problems like the Blurry patches issue.

I guess in the end it all depends on the photo/file inputs the users have.
I believe you that you have no significant problems with the product - I wish I could say the same.


To tom.cass

Honestly, as far as outright bugs making the base program not work, I’ve not had a lot of those. Some crashes here and there and update not working right at all for a long time. Batch process lacks pro features, etc.

My BIG problem has been with the quality of the output. I’m trying to blow up gigapixel size photos a little at a time, which means thousands of photos. Rarely do I have a photo that is a high quality blow up from corner to corner.

As PWC linked to, there’s always lots of very blurry areas right next to sharp areas. This is the number one problem I have in every photo. Often there are hairy rocks, completely wrong leaves, or leaves where there should not be leaves, etc., etc. I have big rants and examples about this elsewhere.

The sharpening is too sharp, like using a large radius setting in PS.

And I won’t even get into the baffling adjustment sections that it is impossible to tell what PAI is really doing what from one setting to the next. I’ve played around with hundreds and hundreds of photos for countless hours trying to make sense of how the sliders interact with each other and I can’t say I’m any closer than I was to making sense of it.

It’s guessing pretty much and see what happens from one photo to the next, because each photo interacts with the sliders differently. In fact, increase the new sharpen 2 enough and you will get large jungle leaves on pine trees! Huh??

Also, I have to blur sharp photos to get PAI to decide it wants to add the right models. If I give it sharp, nice looking photos, it blows up poorly.

The good stuff is, in the parts of each photo where PAI does things right, it is magical and stunning. I was blown away when I first saw what it could do. I would just like some freaking consistency, especially the blurry patch issue.

Therefore, the point of the program is to blow up photos in a quality way. It is most certainly not doing that corner to corner. (To my nature photos. I don’t work with faces.) I work on a 75" 4K monitor and I can clearly see all the problems.

Thus, I want to see the core issues fixed before making the program more complex and likely to have further issues. After those are fixed, then add features all day long for all I care.


In addition, I want feature development to focus on upscaling, noise reduction, and sharpening, rather than general image editing tools.


Agreed. If I want a photo editor, I buy one (which I did, presumably as many of you did). Such a photo editor is part of my workflow, if required. It also already offers object removal, sky replacement and so much more.
A photo editor also has a completely different concept, it is not stateless as Photo AI is right now (load image - do something - save output and forget all on closing), but it requires that the user can create layers, save masks, persist these edited states, reload them - possibly create variants and so on. Nothing I would ever want re-invented into Photo AI.

I bought the Gigapixel/DeNoise/Sharpen AI package as a subscription, which happened to include Photo AI. What I didn’t know at that time was that the end of Gigapixel AI & Co was already reached.

Therefore all I need now is for Photo AI to continue the mission of the previous three products - just combined into one. If that works well (primary bugs like blurry patches, etc. fixed; UI once again capable of handling many files, multi-select + assigning models/values, batch and CLI & established quality control) I have nothing more to complain about.

It’s sad that we all have to spend so much time in the forums here, to keep that process going and to follow up, if it’s now going in the right direction. We all have lives outside of Photo AI, too…

It should be self-evident that consolidation of existing features (fixing major issues & establishing quality control) as well as re-achieving the capabilities that the previous 3 products offered into Photo AI should be the only thing to focus on right now. With over 62 users (in only 2 weeks) conveying the same thoughts I can’t be all wrong…


I would say that the votes received from this post are definitely significant: if I were the company I would not take it lightly but I would read all the comments carefully. I hope that in the meantime the company is at least carrying out a check, even on a sample basis, with those who have purchased the product to understand how satisfied they are with the product purchased and how many are not.


I am slowly losing my patience…


I want to tell you about my experience with TPAI: I started trying the Topaz products with Gigapixel, Denoise and Sharpen more than a year ago: in particular I was very impressed by Gigapixe’s face recovery. I would like to start by saying that at that moment I was, and still am, particularly interested in restoring old prints, many B&W which I had digitized, and in nature photography.
Each package had its own advantages and some limitations, furthermore there were overlaps: for example Denoise also allowed you to sharpen just as Sharpen allowed you to do denoise and Gigapixel did both, but all this, being independent apps that could be purchased separately, made sense .
Then came PhotoAi: I saw it as an integration and rationalization, in a single application of, in my opinion, the four starting functions

  • denoise
  • sharpen
  • enlarge
  • face recovery
    This was not the case: an autopilot was introduced, with often unsatisfactory functioning and therefore requiring manual intervention, but the four basic functions remained substantially the same, that is, independent of each other, always with the possibility of making denoise from sharpen, sharpen by denoise, etc.
    So it seemed to me that in PhotoAI the three previous applications remained substantially the same with a dummy hat built on top called Autopilot: what I expected was that the three apps plus face recovery were strongly integrated with each other so as to become a single app homogeneous.
    In the end I want to draw a veil over the fact that in a year practically nothing has been added/corrected to the basic functions except text processing, and I’m happy about it, and the possibility of specifying in face recovery whether or not to treat hair and neck so now they can be excluded but if they are included the exact same limits as before remain, such as hair where the recovery often only relates to a part of the hair itself (same for the neck).
    What can I say: I’m disappointed. every week I expect an improvement but this never comes.
    Even a release every week is decidedly excessive and gives the idea of a certain neurosis and approximation on the part of the company that develops the TPAI package.
    To company I say: think people, think… before it’s too late!

I’ve stopped updating after 2.02 of PAI.
I only use the export from LRC and I only use denoise and sharpen .
Back in v1.5.4 the RAW model was better that the LRC plugIn for sharpen especially.
I’m tired of downloading a new version to see if it is better than the last in these 2 basic models, and then to be disappointed when it’s the same or worse.
Right now Denoise AI in LRC works really well and is less destructive than TPAI, and the sharpening (not AI) in LRC especially in the masks, is really nice.

By way of example here is an image shot on a Nikon Z9 in HE* RAW at ISO20,000.
It was run through LRC DenoiseAI, the bird masked, exposure lifted by 0.4 and sharpened.


Thanks for sharing your views on the current Photo AI usability; being as grim as they are, that would have been something for Shakespeare… :japanese_ogre:

I think it’s important that Topaz Labs can read all perspectives, not just from my point of view (just because I happened to create this thread). It could help raise awareness. (I know, hope dies last…)

P.S. I’m curious to see what the change history of today’s Sprint release will entail; if it addresses this thread in any way (which is already almost 3 weeks old). :hand_with_index_finger_and_thumb_crossed:


At the moment I would say that this thread has been completely, and I think deliberately, ignored!


Amazing: if this is the result of LRC, it’s a good idea for Topaz to ask themselves some questions and get moving, as suggested in this thread!

I also think bug reporting and tracking using a forum is sub optimal. There should be a dedicated bug tracking website (like Jira/DevOps) where at least beta users have access to. Maybe there is already such a thing, I’ve asked in another thread about it:

1 Like

Sadly there is still insufficient quality control…
The CLI is completely broken in 2.0.7 once again
Error | [CLI] Engine canceled processing

It shouldn’t be too difficult to write a simple test that starts the tpai.exe and checks if at least an output image was created (That’s a 5 minute task for a developer). And running that test prior to every release…

Sadly this is not the first time the CLI was completely broken, there were 1.3.3, 1.4.0, 2.0.0, and so on… It’s like flipping a coin; will it work or not this time. :coin:

I continue to believe that one release a week is something for beta testers and not for a finished product: I don’t think that more than three/four releases can be published in a year, unless exceptional events occur, and after a serious cycle of testing non-regression and quality control of new features


I did not sign up to be a Beta Tester, yet I am being treated like one! :frowning:
It’s very frustrating not knowing what release to get so I can have something reliable and not worry about weirdness rearing its ugly head.

Topaz, please produce a thoroughly tested stable release and then wait a long time before another. And of course I would also like you to adhere to the subject of this thread.

Thanks, Rod