One more observation: if a raw file is imported to PAI and the autopilot selected Strong Raw Noise Reduction, the noise reduction panel has the “Normal”-Model selected. Switching between the normal and strong model has no effect with my raw images (Windows, Olympus OM-1 ORF-RAW-Files).
Dumb question : what the mask is used for ? is it to be used by any model Ai so it apply different work on the “model” and the background, or it’s used when Sharpen is turn on with the option “subject only” ?
just to know if i must take care of redefine the subject when it fails, if it’s not using the sharp model. thanks ! 
2017 iMac, Ventura 13.0.1
Bug:
Using Capture One Pro 21 (v.14.4.1) with the latest version of Photo Ai (v1.1.0) If I move the center image around either before or after adjusting the sliders or using the auto settings, Photo Ai will stop working properly. The progress bar will turn a grey color and show “Viewing Original” next to the progress bar and any adjustments that were applied if any will be undone. Only thing to do is quit without saving.
With my limit testing I found that if I moved the image around within the smaller box in the upper right corner everything worked fine. But as soon as I moved the image around from the main screen it stopped working again. Kinda annoying as I prefer to move the image around to look at it from the center screen image.
I need some more info to debug. In TPAI 1.1, after you replaced the json file, can you do this when you have time.
Close the APP → open the APP → run the test image that you posted (two faces one) with face function on → after processing, click Help: Open Log Folder → drop the most .tzlog file to Submit files.
One could say that a camera on Auto (let’s, for giggles, say that that Auto is the same as AI, or the premise of some AI) will capture what is “right” (to the camera or by whomever programmed & instructed the camera).
That is not the case in many situations! And, it is false to assume it’s true.
A camera doesn’t see like humans do. And, it can’t autonomously make the kinds of creative (or not) choices the wielder of a camera might (near to infinity DoF, selective focus, things that are white on white appear that way, etc.).
The same holds for Auto (AI if you want to be trendy and catch-phrasey & in vogue with what’s currently mkt chic) processing of images. There need to be controllable choices users can make and not be wholly dependent on what a robot thinks is aesthetic or ‘right’.
If aesthetics (which traditionally was Topaz’s playground/purview) is the purpose of their Auto/AI offering then AI can’t be an end all be all & should be a springboard or primarily an idea tester that one may use but most likely will modify in some way. That scenario means that ‘AI’ can’t (& maybe shouldn’t!) always be right (but must be tweakable).
If, perhaps, aesthetics is less of an issue than handling massively bulk output in minimal time than a purely Auto/AI creation with no input by users (or non-discriminating users) is likely optimal. The tradeoff of time vs quality (good enough but not what the user would create if they had the time…) would likely tip to fully Auto/fully AI.
That is a question of target mkt & intent and affects specs & ops of any “AI” software. [Until the fad passes and mkt terminology changes to the next big thing]. 
And, that’s the issue. Why is it “better”? It is a trained robot’s idea of better. And how that engine has been trained.
What if the goal is to create a low key, contrasty, cinematic result for an image.
Then the trained engine decides, "No, that’s not “right” & therefore produces a full spectrum, balanced contrast, possibly more high-key look.
That brightening & AI aesthetics choice will make it tougher for the artist/user to bring the image back to their vision.
If the program is strictly a pre-processor it shouldn’t alter aesthetics of base images with just denoising, sharpening, enlarging.
If users want to not buy other s/w or hosts than, IMO, there should be a panel added with sliders for Brightening/Darkening, Adding Vibrance or Saturation or reducing those elements, etc. And, those settings should be either applicable to a batch or saveable as a preset. The AI shouldn’t take full control of aesthetics.
Is this happening to anyone else? It still happens for me. Photo AI is sitting idle (or it should be idle), but it is using a lot of CPU. The fan comes on. This does not seem to be correct behavior. Why is it using so much CPU when it is just sitting there and not processing a photo? None of my other software does this.
Quantify “a lot of CPU” for the devs … you should be able to get a screenshot of the task manager which will show the amount of CPU and RAM being used.
Note no app that is running stops when it isn’t being used. When you load an application it allocates resources to use and only releasees those when you exit the application.
This is my PC before starting Photo AI. This is typical when I am not doing anything. Note low CPU usage and running at 1.17 GHz. Fan off or so quiet I cannot hear it.
Here it is after I start Photo AI. The CPU that Photo AI uses is generally between 24% and 35% when no photo has been loaded and it is just sitting there. It continues like this until I exit the program.
As you will note, not only is Photo AI, while supposedly idle, using a lot of CPU it has also caused the speed to ramp up to 4.26 GHz which makes the fan come on rather loud. And it will continue this way until I exit the program.
Photo AI will continue to behave this way as long as it is open even though no photo has been loaded. This is not normal behavior for most programs. What is it doing that causes it to use so much CPU and cause the CPU speed to jump way up forever even though it is not processing a photo? It has to be doing something with all that CPU usage (multicores). None of my other software does this.
The application of a raw profile to sensor data really has mostly zero to do with AI, but I don’t have the energy to debate the details further. If you don’t favor the AI approach, then perhaps Topaz apps aren’t for you as their entire focus now is AI. And Photo-AI even more so, as it seems like the target is merging all the AI tools into one app.
However, as mentioned, there are plenty of options for processing raw files that certainly don’t involve AI in the least if that is your goal, some of which also include very effective denoising. If one is interested in tweaking RAW file output for artistic effects you may be forced to make use of one. Raw file data processing has never been a Topaz forte, although it has been improving.
There could also probably be some user options on how the Topaz apps deal with the raw data profiles if you can get the devs interested in hashing all that out; some of it are gray areas open to processing interpretation. My guess is “I doubt it”.
Commonly crashes especially when opening multiple pics. Commonly crashes when only opening 1 pic. I7, 16 GB. I’ve been impressed by 1 image so far after trying multiple. Questioning my decision to purchase. Will continue to try tweaking the settings for better results, but not sure what I can do to solve the crashing on the program.
the cpu temperature rises by 10 °C and the cpu power watt usage increases to 25 watts, but the cpu usage itself is almost nothing.
“Perhaps I’m mistaken, but I thought the entire point of “AI” processing is to have the computer make the decisions on what is “correct” (with the possibility of user override). Isn’t that why we’re using AI in the first place?”
Well… it depends on what YOU mean by “correct” and what the software is supposed to correct. In this case, TPai is only supposed to correct 3 things. NOT anything else… Noise, sharpening, and upscaling…
1 Like
Photo Ai 1.1 is producing some seriously bad artifacts. this occurs a 1x and 2x scale. the beta version was not producing these issues. I tried the same raw images in gixapixel beta latest version, zero artifacts around the badges.
1 Like
Not really; it’s what the camera thought was correct exposure after processing the RAW sensor data.
Did you read this part from the OP?
So, PAI (and Affinity Photo) are applying the RAW profile in a similar manner and getting a similar exposure. DeNoise is not (as Don has pointed out). IOW, PAI & AFP agree, DeNoise differs (on how to do this).
We can debate which RAW profile application is “correct” (my opinion is PAI/AFP). However, applying a RAW profile is not optional. So, your interpretation leaves out one step, which is to get the sensor data to what the camera thinks is the proper exposure, etc. This happens as the first step BEFORE the three potential steps you mention (if at all). Last three are optional, first one is not optional.
Anyone should be able to see that the PAI/AFP exposure is more correct. The example is right there in front of you…comparing those two against each other (ie DeNoise is the odd man out).
Lastly, it may not be obvious, but this has zero to do with the AI processing.
Henry,
My hardware setup is similar to yours, with the exception of missing the Nvidia GPU. My CPU usages also settles into the 30% area when PAI is running after opening without a photo open.
I clicked on Activation again, which didn’t worked at the first time. But the second time it worked well.
Let me re-iterate… If any colors or exposure settings are changing as a result of Topaz’s product’s…Then you need to open a support ticket because that’s an error. Topaz has determined that their focus is as a tool vendor. They intend to provide the best tools for the selected task… Right now, they have 3 tasks that they’re targeting. Shapening, Denoise, and upscaling… If anything else is changing, it’s NOT intended and needs to be reported. Topaz products are NOT intended as a general purpose photo editor within the scope of these these 4 products…
2 Likes
Let me reiterate, it’s not a result of Topaz products per se. It is an issue of what RAW profile that the Topaz app is passing to the third-party Raw libraries. PAI & DeNoise are using different profiles. Call that an error if you wish. As I explained, feel free to have the devs explain it to you.
You apparently don’t fully grasp the concept of a profile being applied to Raw sensor data to get a usable image. In simple terms, DeNoise is sending the Raw processor a different Raw profile than the other two apps (like Don explained, DNAI is using a different profile than the other apps). I hesitate to call this an error for the following reason, but it is a happy accident for what the OP desires.
With camera raw images, there is no single “true” way to render a raw image. Every application renders raw files differently, and every program allows the user to make different adjustments.
As a Beta tester, you should realize you have THREE examples of end result on which to compare. TWO of them agree with each other (PAI & Affinity Photo) and the third (Topaz) is basically not doing the correct thing (DeNoise). This is an issue that has been known for months/years.
I don’t mean to be rude, but you sound overly confident for someone that doesn’t appear to understand what the issue is.
1 Like
To put it bluntly, you are under-informed.
Quote (from link below)
The brightness information in a raw file is recorded very differently than the way it is stored in a standard digital image. If you were to display the raw file as a standard image, it would be very dark and low contrast.
This link does a decent job of explaining the Raw transformation.
