PhotoAI not upscaling correclty

It’s impossible to upscale an image correctly using the combination of size (cm) and PPI. The only way to correctly upscale is to enter the width (or height) in pixel.
Moreover, If I enter a size in cm, I’m forced to use a resolution in px/cm instead of keeping px/in that , unfortunately, it’s a standard.
It should work more accurately how Gigapixel was working (at least for the upscaling part).

Steps to reproduce issue:

  1. Open an image you want to upscale (TIF file)
  2. Set the W and H dimension in IN without changing any value
  3. Set the resolution to 360 px/in
  4. Try to dial 90cm in W dimension. PhotoAi will go crazy and will not work. It will not let you to go that size)
  5. Close PhotoAI and do it again by simply entering 12756 px in W. Not it will just work

Honestly I feel this is a massive step back from Gigapixel AI.I also don’t understand why the basic way to work is upscaling by a magnification factor, as who on earth would magnify by magnification factor and not by real-world size.
Moreover I don’t understand why there is still no plugin for upscaling like Gigapixel (even if Gigapixel was bugged until the last release 6.6.3 forcing the sRGB conversion every time, so I hope that once released, the color space conversion bug on PhotoAI will be fixed)


Topaz Photo AI [v1.5.3] on [Mac]

As per GigaPixel the maximum image size is 32000 x 32000 pixels.

You cannot set both width and height otherwise the image would be stretched.

Choose in or cm and not pixels for either width or height then the ppcm or ppi to scale.

1 Like

Hi Don,
Thanks for the reply.
The maximum allowed size is fine: 32000 pixels at 360 DPI will result in a 225.77cm print (88.88 in).
The issue is that PhotoAI is not able to manage conversions: I’m not able to get evean a 90cm 360 DPI image.
If you try to recreate the bug with my sequence, you’ll see that PhotoAI lose track of what is doing because it seems it’s not working on reversible paths and related values.
Just to give you an example, if you resize the image on Ps, you’ll see that once you change ONE parameter between W, H and DPI, the other two are correctly updated. Then if you change again one of the other two parameters, the one set before and the other one are updated again.
Not in PhotoAI, where after the first input, if you want to change something else PhotoAi is not able to update conversions and keep track of the math.
Moreover again is not possible to mix W and H cm sizes with PPI values, which should be really the basic (PPCM doesn’t exist in real world)

That is what I tried to explain recently. Images have a fixed size in pixels, not in dpi, ppi or lpi because those are related to the output device, not to the image. :neutral_face:

That is not possible. How many dots/lines/pixels PER INCH (or cm) you need to get the image to a certain width/height in inch or cm ON PAPER (or on screen) depends on your output device, will say YOUR MONITOR or PRINTER. The image and/or application may not know anything about your output devices and therefore can not calculate any dpi/ppi/lpi needed to deliver a certain size in inches or centimeters on paper or on sceeen or anywhere …

1 Like

You cannot mix and match, you either have inches/pixels per Inch or centimeters/pixels per cm

The pixels per cm and pixels per inch are used to resize the image and not to print the image. It assumes that you want a image size, in pixels, to fit in the length/width specified in pixels per inch/centimeter.

1 Like

Hello guys, thanks for the reply, really appreciated.
I know the differences between PPI,DPI, output devices and things like that, sorry for my oversimplifications to go straight to the point.

What I wanted to say and that I confirm is that PhotoAI is not able to make correctly conversions.
If you follow just the steps as the bug subbission requires you’ll see that either you use basic conversions (magnifications, single single dimension), or PhotoAI doesn’t work.

Even Gigapixel was working on that, and the only reason I’m not using it is that even at version 6.3.3 nobody notice and fixed a major bug like the color profile conversion.
PhotoAI is for sure a promising application, but so far to me it looks more like a Luminar style software than a professional and reliable tool like it was before.

1 Like

A good summation…

1 Like

Sorry, why is this topic moved out of bug section?

We are releasing this plugin for Topaz Photo AI in two weeks. It will be in the File > Automate menu for Photoshop.

Why do you use both inches and centimeters? Why not set it to 35.4 inches @ 360 DPI?

Topaz Photo AI is currently set to use all inches or all centimeters for simplicity and this is the first time I am hearing of a use case for both inches and cm. It seems very rare.

Hi Lingyu, and thanks for the reply.

Great for the plug-in. Please double-check that it is not affected like Gigapixel by the colour space bug, so that when the upscaled image is set as a new layer in PS, the image is it not converted automatically to sRGB color space from a larger colour space (like AbobeRGB or ProPhoto RGB).

The general reason behind the conversion request is that when you want to print something, of course you need to think according to the final output and the different requirements, which usually are the size you want to print and the resolution of the printer.

The first parameter, the size of the print, is usually dependent on where you live. As I live in Europe, we use mks system and so I need to refer to cm as size. But if I live in US I would probably use inches.
On the other side, the second parameter (the resolution) is mainly defined by "what you want to get "(a super simplification not to take into account viewing distance, limits of the mages and many other bla bla bla) and who is physically printing (so let’s say the printer, in terms of the machine). But that second parameter, so the resolution, is worldwide defined in inches, never in cm.

So, for example, it’s quite common for me to receive a request for a print in a cm size with a dpi requirement. In many years of printing, I never commercially got a request of pixel for cm, and if you check all the printer specs, the max resolution is always expressed in dpi too (regardless EU or US)

Right now, due to the upscaling square bug of Photo AI, I’m using the standalone version of Gigapixel (that is not affected by color space bug) to prepare an exhibit, and there I had the chance to move from a size in cm and a resolution in inches with no issue like it happens when you resize an image through Ps (so very good). With Photo AI (before finding the upscaling squares bug) the only way to get the right size was to calculate offline how many pixels I need for the specific output (for example, for my 90cm on the long edge at 360dpi, I needed 12756 pixels) and then telling Photo Ai the exact amount of pixel I want.

Sorry for the long message and I hope it makes sense!

1 Like

To be honest it is the job of that printing service to calculate everything right according to their needs to produce the print. it is not your job. If that company is not willing to do their work I would just switch to a better service! :person_shrugging:

This does make sense. However, we would likely receive complaints about changing the resolution for CM to DPI.

I would recommend converting CM to IN. The input for width/height has up to 3 decimal places so it will have enough precision for your prints.

If there are other users who could weigh in on using DPI for both Inches and CM, I would be interested in hearing from them. One data point is not enough confidence to make this decision.

Thanks for the reply.
I understand your point of view, but honestly, it is not a professional solution, and no, it’s not my task to make a conversion, it’s the task of a professional software. (and again, it’s a ridiculous basic function)

You could just add the “arrow menu” (sorry I don’t know how to call it) where you can switch from IN to CM. So if you start in IN you find DPI and if you start in CM you’ll find “DPCM”, but then you can switch from one to another (actually, like you have already in Gigapixel, so nothing new)

I mean, to me, it would look really a surprise that somebody among users uses a “DPCM” that really doesn’t exist.

Again, I bought and I’m using Topaz apps as professional tools, but recently I have more the feeling you want to create something close to Skyum Luminar to target another layer of users, the ones that don’t want to learn anything, that believe that one button is the solution of everything.
Absolutely fine and I understand from a business point of view, but honestly I would never ever choose a UI option according to “let’s see what other users think”. In the commercial/professional world there are standards, that are standards set for a reason, and they are not changed by a poll.

And to reply to Imo, ABSOLUTELY NOT, it’s your job, not the printer’s job, because its’ your need (!!), and the first rule of a good print is that you prepare the print. If you send an undersized image to a printer, if you set the wrong color space, if you don’t know how to proof, then then the print will be 101% a s**t, and that’s only your fault, not the printing service.

Again, super sorry if this thread looks polemic, definitely not my intent, but honestly I really feel I’m out of place.

Thanks for everything

I sent you an email as I’d like to learn more about this topic. I am not a print professional, nor is anyone else on my team.

The feature you are requesting is simple. Even then I need to verify if a feature request makes sense with data and multiple points of view if I can get them especially because I am not knowledgeable on this topic.

We have users who want a simple solution, and users who want full control. We do not listen exclusively to either user. Most of our users are photographers, but there are many who print their images. Topaz Photo AI is not a professional print software but we would like to build it to support those functions.

You want to upscale an image to be printed with 300 dpi? 300 dpi will not be enough if for example you want to print with grayscales in 64 shades for example. In this case the machine which might be able to print only black and “white” needs 8x8 printing pixels for one of your image pixels to get 64 shades of gray for a pixel. In this case your 300 dpi image will be printed with 300/8 so maybe 37,5 dpi. :eyes: and of course it is the fault of the printing service. We once asked a local printing service why all our images were wrong until they told us that they don’t even work with the included color profiles and that they use their own profiles to print images …

Sorry Imo, but I believe you should read/study a bit more about printing, and maybe it’s you who should change the printing service.

In any case, this was not a topic on how to print an image but to report an issue with Photo AI, and now we’re going out of topic.

For me this thread can be closed.

Thanks Lingyu, got your mail, I’ll reply there.

1 Like

When you want to drive via Taxi and the Taxi service requests the exact driving distance in miles per hour you would suffer from the same issue because the miles per hour refer to the driving speed of the taxi, not to the length of the stage.