Hi, guys! Here I am with one more discovered problem, namely a limitation of the dilation settings in the export image and data outputs dialogue window, where we can set only one percent color expansion beyond the boundaries of the UV islands. We need more, a tad more than just that one percent! Here’s my proof:
As you see there are no problems with any color bleeding in the Instamat studio. But when the textures are imported into UEFN we can see the color bleeding as a thin white strip:
Witness with your own eyes! The guilty edges are too close to the washed-out boundary between the dark green and white colors. And yes, I tried to crank up the diffusion to its maximum as well but the result was the same - ̶I̶’̶m̶ ̶s̶t̶i̶l̶l̶ ̶n̶o̶t̶ ̶a̶ ̶B̶r̶i̶t̶i̶s̶h̶ ̶m̶o̶n̶a̶r̶c̶h̶ the strip is on its place.
Hi again! I hope you were SIGGRAPHing to the fullest. While Adobe is making mistakes, it’s your time to take over the world. As for the noticed problem, let me explain more accurately that it is mainly related to the fact that even if the diffusion property is set to disabled or enabled with zero intensity, we have that pretty heavily diffused boundary between different islands’ colors and that is the reason of the bleeding. I know that the compression UE applies also plays its role but that always-presented diffusion is the main culprit. So you could let us regulate the dilation to shift the washed-out colors farther from the edges or solve the problem by reducing the diffusion to its feasible minimum when it’s turned off. Of course, I can flesh out the dim colors manually in GIMP or shift a bit the affected edges in a UV editor but the thought of fixing tons of UVs or textures makes me feel slightly uncomfortable.
Based on your screenshot of the UV layout, the shells look like they are packed very close together. I’d recommend what @voidberg mentioned earlier, which is to increase the spacing/margin between tiles to see if this helps.
Guys, I have good news - Instamat is completely innocent of what I suspected it of! I conducted my own thorough investigation with the involvement of at least two well-known British scientists who asked me not to reveal their names due to their unbelievable modesty, and this is what we were able to find out:
When we use the dilation of 1, the color boundaries are completely crisp as they should be (the map is opened in Blender):
When we use the dilation of 1 and enabled diffusion of 1, the boundaries are partly crisp and partly diffused and that’s ok because I’m never ever going to use any diffusion (the map is opened in Blender):
And here is the source of my problem - the default Mip LOD bias value in UEFN. It is set to 1 not 0, and with its resolution and compression the texture gets those nasty washed-out color boundaries (the map is opened in UEFN):
With the value of 0 for the Mip LOD bias, the model has no bleeding at all, though the default compression blurs the borders a bit (the map is opened in UEFN):
Now I find myself in a situation where I can’t change the bias due to performance considerations and I’ll have to shift the affected edges to get what I want. Anyway, Instamat has nothing to do with the issue. So just accept my sincere apology for my malicious attempts to compromise our awesome program. For too long I seemed to be on the dark side!
Sure thing! And thank you guys for your passionate involvement in the investigation. Now we have every reason to establish our own digital detective bureau.
