Quixel Help Center home page

Curvature mask fails to detect seam edges on custom meshes

Answered

Comments

11 comments

  • Could we see this model and the results of it in the viewport? Would you be willing to upload it so we can look into it?

    0
    Comment actions Permalink
  • CraftyMutt

    Here's my curvature mask result in mixer side by side with the original model in blender, it is a fairly basic model as this was intended to be a first quick test

    You can see with these side by side that the seam edges (in red) on the model do not show up in white on the curvature map, which is a shame because I was so interested in trying Mixer to quickly and easily add scratches and such along edges of assets.

    I would be happy to upload the model (and material ID mapping) if this can help investigate what's going on! Is there some file attachment button I am missing on this page or is there another way this is done?

    0
    Comment actions Permalink
  • You can upload it to Filedropper.com. I'm currently transitioning over to Cary, NC, so I can only work with 3D files in a limited capacity at the moment. By next week I'll be set up and able to help more with model troubleshooting, although we only do this in select circumstances.

    That being said, I think you'd be better off adding some edge chamfering to this model or adding a baked normal map for Mixer to generate curvature from. There's not a whole lot of data to work with in what you've provided.

    0
    Comment actions Permalink
  • CraftyMutt

    I've uploaded my model here

    I will also continue to experiment on it with your advice. I had hoped that once I'm familiar I would be able to use mixer to mange and make textures for both realistic and low poly style game projects to avoid having to work with too many different programs at once, but it sounds like perhaps Mixer is not well suited for working with very simple low poly models? It is obviously working great with highly realistic assets such as all the ones found in the library!

    0
    Comment actions Permalink
  • oPryzeLP

    Any updates? Having the exact same problem

    1
    Comment actions Permalink
  • CraftyMutt

    Unfortunately after investigating several quixel megascans assest in my own time and with no feedback from the quixel team, I have come to the conclusion that it is a simple technical limitation (or unnoticed bug) in the software. I did learn all by myself that it is not the marking of seams that causes the behaviour, but the parts of the UV map being separated when mapping - this happens naturally when marking a seam, but also cannot be avoided just by not marking seams and manually separating the parts of the UV map.
    It certainly would be nice if someone on the quixel team was aware of this limitation so they could quickly and easily answer this question each time it comes up rather than putting users through some song and dance of providing information to allow them to investigate the problem, only to never reply again! Even better if they knew and passed it on to software development, maybe someday they'll be able to make it calculate edge angles regardless of whether two adjoining parts of the model are separated in the UV map.
    I do genuinely hope it'll change someday :)

    0
    Comment actions Permalink
  • oPryzeLP

    Thanks for the follow up! What's weird to me is that some assets I've seen on bridge appear to me like they would suffer the same problems, however, miraculously they're fine. Daggers specifically, have perfect curvature even though their UV maps are separated along the sharp edges. There's gotta be some sort of trick to it that we're not seeing.

     

    Regardless, I'm stunned as to how this hasn't been brought up more.

    0
    Comment actions Permalink
  • CraftyMutt

    I must admit I didn't check any particularly sharp objects, but I did notice in for example a wooden crate that the corners mostly appeared to calculate edges as expected, but on close inspection you could see where the UV edges were since there was a slight discontinuation in the mask, but due to being extremely high poly scanned assets there are so many polygons around the corners that they disguise the join - the more polygons, the gentler each one is affected by the curve individually and the more the software knows is going on around an edge, the less obvious it is that data is missing from the other side of a split. By making my own meshes with carefully planned maps, making splits at the flattest possible places and adding extra geometry around edges where I need to put a seam to create several gentler slopes and smaller polygons for the software to calculate and colour with I can get it to somewhat disguise it the same way. It does not help so well if you want to use the software for relatively low poly assets directly, though I am trying to get by without using curvature masks in those cases. Gradient masks and others are still very useful!

    0
    Comment actions Permalink
  • oPryzeLP

    Yep, I noticed this as well. I added more polygons around the corners and carefully placed my seams to achieve, more or less, what I was going for. All in all, it's a weird problem to skip out on. I won't pretend to know what's going on, but given how amazing Quixel is, I'd think they'd at least have a checkbox for detecting normal changes along seams.

    0
    Comment actions Permalink
  • Oleg Antipov

    I faced the same problem: the curvature mask is simply not calculated where a mesh has UV seams. This problem is visible even on 3d models from the library (however, due to the high poly, this is not particularly noticeable: https://drive.google.com/file/d/16WuSl34Zi6rrirE4cEL0IrKW4PtVNEXh/view?usp=sharing 

    For low poly models, this is generally fatal: https://drive.google.com/file/d/1kqXpnrOak80AQ07ZOiDiUnzYQmda8Esy/view?usp=sharing 

    I use Blender, and it has the awesome TexTools plugin that can generate the correct edge mask (as well as AO, material ID etc.) and then use it in the Mixer as a custom Image Mask: https://drive.google.com/file/d/1AqQoRmW_cb6XN7Gm1O8x9yrBJex5KJUQ/view?usp=sharing 

    1
    Comment actions Permalink
  • Jonathan (Community Manager)

    Hi Oleg Antipov - we're aware of this and will be looking into improvements in this regard in a future update to the software. In the meantime, I'll be closing this thread. Thanks for your patience!

    -2
    Comment actions Permalink

Post is closed for comments.