Hey @TyXanders,
Apologies for not pinging you on my last message, I’m not sure if you got the notification about my reply
Ah thank you! @ath0rus checking it out now, and will raise this again to the team just in case.
couple of things came back to me reading through this.
- there was discussion about the pixel difference being minor
- the low presences of very low error margins.
- different terrains different error margins
led me to think that that as tiny an error margin is the scale and size coule potentiallt exacerbate errors
i noticed you’re already working in a 0-1 scale anyway, and i tested true scale already came to the same results.
the only thing left would be to overlap the error and blend it.
for this case i had some really good initial results with the following build set up for a tiled build type:
I really hope this solves it for you.
Hey @TyXanders,
Sorry for the late reply, Been a busy few nights on several other projects.
Did some testing with the settings you suggested (1 pixel overlap) and that gets it alot closer and removes the big gaps, While there are no big gaps there is for some reason still a seam (see the pic below with an excessive amount of arrows), It’s think the geo is refusing to coordinate.
Since I think this is now at the point where It’s a blender issue not Gaea I will probably submit an issue report to blender themselves and see what they have to say. I also tested with a stupidly high subdivision and that changed nothing, Still both types of seams.
Is there anything I should test before I report the issue to Blender to see what they have to say?
@ath0rus no pressure on time,
Alrighty, this could show us how Blender interpolates data and perhaps even auto stitches ( or lack of)
It may be blender-specific, terrain-specific ( or both)
What looks to be happening is that Blender blends the overlap cleanly using the two data sets it might be a default smooth lerp or something else; in my case, I may have just had the best pixel overlap/blending set up for the terrain size I had, incidentally, yours could be different.
In my team’s pipeline, we created a separate tool that blends the overlap for us because the engine doesn’t. Even at extreme angles and multiple points, it does end up clean.
IMO Gaea does create the right data it’s the interpolating data on a renderer that needs to ingest something.
In Blender’s case, it looks like the right combo of pixel overlap and blending from Gaea.


The two things I would check are
- different pixel overlaps
- blending %
Compare the best results
Also, check with a texture to ensure it’s receiving well at the untextured seam. The view may be normal when untextured.
You may already be exactly where you need to be; it’s just finding the right combo.
Hey @TyXanders,
Did some a few days ago with different pixel overlaps
I am 99% sure these are 50% overlap
1 Pixel overlap
2 Pixel overlap (causes a seam)
4 Pixel overlap (broken)
25% overlap 1 Pixel overlap (looks near identical to 50% 1 pixel)
For validation 50% Blending 1 pixel overlap
100% Blending 1 pixel overlap
0% Blending 1 pixel overlap
Looks like at 1 pixel overlap 25/50/100% has no change. Looks like the best so far. Though still not sure what to do to get rid of the extra edge….
Hey @TyXanders,
Finally got around to submitting a Blender bug report to see what they have to say, Here is the link if you want to follow it.
https://projects.blender.org/blender/blender/issues/160551
Hey @TyXanders,
Back again with some more testing, I am now getting seams in Gaea from the tiles I exported (same images from my last post). Same way as Blender was having
Same issue present regardless of the blending percentage, This was with 1 pixel overlap.
Just to note these were very subtle and I had to zoom pretty far in to see it. Sharing as seems to be export issue
Did you test 0 pixel overlap?
Hey @ATom,
yes, see my first images, Right up until Ty told me about the 1 pixel overlap in this thread I was using 0 pixel overlap
As Tom mentioned, this is likely the 1px overlap you’re seeing on Gaea’s side.
Looking at the Blender thread, assuming this is the same file that’s what you showed them.
The user Martijn Versteegh’s last post mentioned that the seams already existing could point to pixel overlap, or what I mentioned previously about the error margins
Min 0.000000023
Max 0.000000976
That’s hard to tell.
If you switch back to a 25% blend with 0px overlap (build defaults), that would be a cleaner file version for them to dig into, which might be good for at least another perspective.
looking further into this!
I ran this in unty too where i did have gaps and had to run our stitcher.

Hey @TyXanders,
Apologies for the late reply, been busy with family and study again. Did some testing with 0 pixel overlap and 25% blending (4k total with 2x2 2k tiles) and I get the seams again
That image is at a scale of 1.0, This next image is at a scale of 2.0 to show the seam clearly
Bumping up to 8k total with 4k tiles and still the same 25% blending with 0 pixel overlap
Seams are far less visible but still present as seen in this image
Interesting thing to note, there still appears to be a 1 pixel border around the tiles regardless of pixel overlap
The last image to me kinda indicates the issue may be on blenders end as there is no seams present in Gaea when imported back in.
Not sure I can re open the Blender report I made due to the time since it was closed, but if needed I can open a follow up.
Out of curiosity, Are you able to share the stitcher tool you are using inside of unity, I may be able to pull it apart and make a Blender version (which I will share if I make it work).
I am writing this to share my findings so far, Still annoyed about the seams and that there is no clear fix, but trying not to complain much about it
Hey @TyXanders any updates on my last comment, Apologies for such a late follow up, I have been reallly busy (hopefully that finishes soon to a degree)
Ooo, thanks for getting back I hope busy is a good busy!
I haven’t heard anything just yet. With all our focus on G3 and G2 updates. I will put it back in after early access settles a bit more see what we can do.
Hey @TyXanders ,
Half update, half info for anyone else running into this. (short version at the end)
I have been poking at tile builds a bit more whenever I get the time, and I think I have tracked down the seam issue (thanks to help from a blender pro). It looks like it’s tied to the resolution of the tile the heightmap is applied to. When I apply the heightmap to a grid object (just a plain plane) with 2048 verts per side (matching the texture resolution) the seam goes away entirely (at least as far as I can see).
The catch, at that level of detail each tile ends up around 4.2M verts, which is brutal for performance. So the logical next step is to figure out a way to stitch the edges of the plane tiles together (or combine them all into one bigger plane and do the stitching on that single mesh (maybe via a UV map)) so I can keep the geometry low while still killing the seam.
On that note: Ty, would you be able to share the addon / code for the stitching method you showed off a while back in UE (the GIF)? I’d love to see if I can use it in Blender or at least learn how it works so I could build a Blender version of it.
Separately, on that 1 pixel oddity around the edge, it might be a genuine bug in Blender (see this report). I have been talking to a Blender pro about it and how they’ve been working around it, If I have any info I will share here if I can.
Short version
- Simple fix for now: use a plane / grid with the same resolution (verts per side, or X/Y subdivisions) as your tile heightmap.
- Next step: find or build a tool that stitches tile edges, like the one Ty showed in that GIF.
WOO progress! that was a happy read for you to find some significant clues and results.
I suppose that also makes sense; as in a heightfield-to-terrain system in real-time engines, you’re often deploying an n+1 resolution of heightfield data, the physical scale retains some level of flexibility for art direction or to hit performance/design targets.
I have marked this for me to follow-up on, as our stitcher is internal tooling; I may be the boss, obviously, but that still doesn’t let me to break policies and licences ![]()
If I can’t share the codebase, I may be able to explain it conceptually instead ( we don’t have patents, so that should be fine)
Thanks for the continued updates!






















