the effect becomes even “worse” when you shrink the application window to the smallest possible size. then you get incredible detailed previews. don’t expect the final processed image to look even close to the preview. it will come closer to the blurry original one. 
Eric Lowery is currently investigating this issue for us. Please rise a bug report ticket about this. Eric said I am the only one who has reported this issue. 
Report here: Topaz Labs
1 Like
I hope they will fix this soon. Is an important bug
https://i.imgur.com/t7uc551.jpg
I tried 4.4.5 and still the same bug. I have never noticed this blurry bug till today.
I know everybody’s system, and the types of photos being processed, are all different, but I’ve tried to replicate the issue you’re reporting using some of my photos with no success whatsoever.
The out-of-focus areas of my images stay appropriately out-of-focus, and the in-focus areas stay nice and sharp. The preview area in either case looks just like what the finished product ends up looking like when viewed at the same scale.
Your examples clearly don’t show that happening as your preview appears to be putting everything into much better focus than both the original, and the resulting enlargement.
From my perspective the in-focus areas should gain resolution and stay in sharp focus. The out-of-focus areas should stay blurred, with the additional resolution providing more smoothing.
Maximum scale multiplier is now based on max pixel count of 32k (instead of just 6x).
Unfortunately this still does not work in 4.7, nor in 4.6 or any version after 4.2.2.
I hadn’t checked this before, but it looks like you’re right about it being limited to 32K in the largest dimension. (up to 30X enlargement on a 1050 x 700 jpg image)
The other thing I noticed was that anything above 6x looks significantly blurrier than 6x does. I confirmed this by adjusting the magnification so the different resolutions are displayed at the same screen size.
Additionally, if you do this in two steps such as 6x and then process that enlargement at 2x it stays much sharper than doing it all in one 12x step.
The 32k px limit is imposed by the image formats (TIFF, PNG, JPEG). The 6x limit (aka extreme blur above 6x) is in place since after version 4.2.2, which is why I keep 4.2.2 installed in parallel to current versions.
On a side note: It is still not possible to enter non integer multipliers (like x4.3) and the maximum multiplier is only x7 even when the result is smaller than 32k px in any dimension.
@timur.born,
sorry this isn’t through. Must be a fault with your installation.
Which country and language setting do you use in windows?
Proof on my installation:
Use 1.6 as multiplier.
Multiplier 19
The 32K limit is from file format and not Topaz fault.
The error with non integer multipliers, was fixed in 4.1, has a open ticket to this bug and confirm it closed.
Again, something I hadn’t tried before, but the non-integer scale factors do work. At 6.00x and below they are nice and clear, above 6.00 to the upper limit they are quite blurry.
Interesting. I asked support about the integer/decimal point limitation many months ago. I am using English language, but German number/date settings. In Germany we use the comma “,” as “decimal point”, but GP ignores this = nothing happens when you enter something like 1,6.
“No problem, just use a point then”:
But when I try to enter a decimal point “.” then I cannot enter any number after the point! So I cannot enter 1.6, because after the “1.” I cannot enter any more numbers. As a consequence I can only enter full integer numbers. It has been this way since I started using GP and I never got any suggestion from support how to fix this or even that it is indeed working for other users.
That is unless I try numbers below 1.0, then I can enter numbers after the decimal point, so 0.8 can be entered. Except when I am trying my very specific large test image that is, with this image when I try to enter 0.8 or 0.9 it reverts back to the last number I used before that. 0.2 - 0.6 do work, though. Strange stuff, reported months ago, got no questions back about it.
In any case I can only enter one single digit after the decimal point, while usually at least two digits are needed. I always worked around the limitations by using width/height instead, but this involves more math on my side (or more entering the numbers and reading the results in Photoshop first).
Concerning the maximum multiplier: My perceived maximum of x7 resulted from a specific test image where GP listed x7 as maximum despite the target resolution only being 29400 px on the long side. I tried another image and GP offered x44 as maximum with a resolution of 31680 px on the long side.
So the x7 limitation I am seeing with the large image seems to be based on the maximum being the largest possible integer below 32k px, not the maximum being 32k px!
PS: How did you put images in your post. When I tried this (either via IMG link or via upload] I was told that I cannot put images in my post?!
@timur.born
Strange
I also use English with German date/time settings.
The problem not able to enter digits after the point was the same problem I had before 4.1 update.
I also can’t enter two digits after the point, but don’t know if this is a bug or a feature.
Before the non integer input worked I also use weight and height. Still use it regular, the fractional scaling isn’t very useful for getting defined picture size, use it only if exact size isn’t needed.
Yes the limitation depends o the picture and is shown In the brackets of the Custom line under the Scale buttons.
To add the picture I use windows tools (Snipping or the new windows 10 one).
Press Control C to copy and paste them in the text with Control V. They are automatically upload and inserted. I use MS Edge as browser.
Bye oder auch Bis demnächst und evtl. mal 4.7.1 komplett neu installieren.
eGPU support on the mac version still appears to be broken - it last worked in 4.5. When I force the app to use the eGPU, the interface graphics appear messed up and it will freeze when attempting to process any images - probably because the application isn’t recognizing the eGPU. I’m on a late 2013 15" Macbook pro, using a Radeon RX 570 eGPU. When I open up the graphics info tab, it now says it’s using the Apple software renderer, as opposed to actually seeing the GPU under version 4.5.
It does work fine if I allow it to use my internal GPU at least.
1 Like
I like this software very much, but there are some troublesome strange image noise problems, this is a good example: http://119.82.26.221/example2/
After you post 10 comments, you will be able to upload images in your posts.
Hi there, do you have face refinement turned on when the crash occurs?
Yes, it happens when face refinement is on.
Are you guys aware, that from what I can tell, if you’re using AUTO mode for a batch of images, Gigapixel AI uses only the FIRST image in a batch of images to determine the AUTO settings for all of the images in that batch?
So, for instance, if you have 10 images in a batch, only the FIRST one will have the correct, best auto settings for that image. All other images in that batch will not have the best AUTO settings, but use the settings of the first image.
You can test this easily by looking for repeatablity.
If you put a single test jpeg image into Gigapixel AI (again, in AUTO mode), you will get the same exact jpeg output results every time you drag that single test image in by itself - the same exact filesize, the same exact result (obviously assuming all other settings remain the same). So if you do that 5 times, you’ll get the same exact results 5 times… or 10 times or 20 or whatever. It’s repeatable.
Let’s say you take that same jpeg image you used above for the single image test, and now you make that image the 5th in a batch of other images, (again, in AUTO mode), you will suddenly get DIFFERENT results for that image. Sometimes VERY different results, all depending on the AI analysis of the first image of the batch. You will know this because along with probable visual differences, the output filesize will now be different.
Obviously, this is not good. When in AUTO mode, each image should be evalulated individually regardless of it’s position in a batch. And regardless of it’s position in a batch, in AUTO mode, it should have repeatable results every time.
This is currently not the case.
Please test. You will see that I am correct.
Also, this may have been brought up, but you can no longer drag and drop a folder into Gigapixel AI. It doesn’t work and now returns an error.
1 Like
Please open a bug report ticket about this.
Hi @taylor.bishop
Seeing your contribution to another member re: crashing with Face Refinement turned…I posted this earlier in the thread. I hope it will be a pointer of interest in regard to what I experienced a version or two back and was cured when I tested another version provided by Ashley Tharp at the time.
Can I refer you to those previous discussions 
HTH 
你不是唯一一个，我也是如此，已经把所有版本都安装了一遍……极度沮丧，预览简直完美，保存就如狗屎
You are not the only one, and so am I. I have installed all the versions again …
Preview is perfect, but the saved image is not the same as the preview.
