There is a work around for this, tile the pattern into a 3x3 with your image tool, then topaz enhance and center crop. This works very well in any filtering/enhancing tool. It will only fail when the algos are not predictable and vary with result on the same repeated image, this can happen with the random patchy blur vs. details issue that topaz photo ai seems to have, but it is far better than the warped seam in practice.
You can also tile to 2x2, then enhance, then center crop, then image shift the corners to the center.
Of course this wastes time/memory and it would be far better for the tool to figure out how much of an edge wrap around buffer it needs for the algos it uses and do this transparently. So backing you up on the request - for both photo ai and gigapixel ai. Be sure to have x tile, y tile and xy tile seaming options.
I use imagemagick/tpai CLI so it is easy to pre/post process in a .bat file. You just need disk space and cpu/gpu time! The only problem with tpai is it uses the GUI auto preferences only, no controls.
The 3x3 above is the way to do it. It happens because the process looks at all pixels around a pixel and when it gets to edges there are no pixels there… so it guesses. Going to a 3x3 not only places pixels there… but it places the right pixels next to the center box.
I thought that is what the 1:1 option did? Had not used it yet but looked at it but since imagemagick cropping is working and did not seem if I could batch tpai crop I did not pursue it.
I settled for 2x2 because of 4x the time/space instead of 9x with 3x3, though 3x3 is an easier crop. 2x2 does mean that seam issues show up in the center, 3x3 those issues can appear at the edges so either way is a problem. Low-Rez enhance has issue with cartoony streaks and patchy blurry vs. details that can cause seam issues. If it does not have those issues even when the original texture was poorly seamed it does a great job fixing it. You might need to run a Standard enhance pass since it has the compression fixer while intended for jpeg blockyness seems to work on bad source seams when you pretile. I am doing the 20yr old morrowind RPG and it has some really bad seams!
if QA is willing to spend $15 (or wait for frequent discounts)
includes a lot of seamless textures as uncompressed unarchived loose files for modding purposes, and I can point out the worst ones that 1x1 has problems, as well as 3x3 or 2x2 pretiling still has ai model issues.
I’m sure there will be a more efficient way for topaz to fix this, but there is no reason that there couldn’t be a “tiling” option in settings where this 3x3 is created then cropped behind the scenes without the need to expose the process to the end user.
The hack works, but it’s a bit of a time suck at the moment.
The crop feature itself could do with a little UX enhancement too TBH.
That partial solves the problem. 3x3 tile on import then center crop output and work on a temp image, which is very important because 3x3 batch mode preprocess tile, post process crop takes 9x the input and output space.
They both take the same amount of topaz AI time, so optimizing that means the devs figuring out how much buffer is needed, and also making it a wrap around buffer so that the exact same solution where it meets on both sides of the tile edge is done.
But just having the quick and easy internal 3x3 (or 2x2 for less time/space) is a big win on space that I will take, I either play games or enhance game textures - do not have room for both!
The first step in that is to add a automatic 2x2 recenter crop and a 3x3 center crop option (with prefs preset), currently it is an aspect choice list with manual crop moving. Impossible to use it for tile crops
2x2 tiling then recentering crop makes the edges seamless but then the centerline might have AI mismatches.
3x3 tiling then center crop the edges might have AI mismatches but no centerline mismatches.
If the input texture has seaming errors to begin with it does a great job repairing it, then just might add its own problems.
It depends how blurry/detailed the edges are, and the problem is the AI is not consistent across repeated sub images. There is a known issue with no solution, they have inconsistent blur/detail patches across an image that changes with AI and parameters.
Both ways eliminate the boundary edge warping so it is significant improvement.
Only devs making every ai do wraparound so blur/detail exactly matches on edges will truly make it seamless.
