It looks very nice.
I hope you're going to try to fix the issue with the bleeding around the edges of surfaces, like at the bottom of the pillars in the above examples.
Hm... But now that I take a second look at it, it looks like both versions have the same lightmapping issue.
And if I am not mistake, that second level is a maplet level?
So maybe pert or all of the issue is with how maplet sets up it's lightmapping coordinates.
But I dunno about that, cause there's no way to make lightmap coordinates line up perfectly with all the texels. So all games must have to make corrections for this issue during the lighting calculations themselves.
I wonder if one could solve this by calcualting a lightmap with 2-4x the detail level, and them averaging it down. That then would make it so that if a texel is mostly outside of an object, 4x as many pixels might be lit as unlit, and the averaging of them down to a single pixel color would result in a drastically reduced darkening effect on those pixels where the center of the texel is inside an object.
But I'm not sure that would entirely solve the issue. Maybe just reduce it by half.
I think what you might need to do is sample the 4 corners of each texel, and if any of those is exposed to the light, then the pixel is calculated as if it is in light.
Or you might be able to calcualte the lightmap at 2x it's width/height, size, and then take the maximum pixel color in each 4 pixel square when downsampling to a single pixel, instead of averaging them.
I hope you're going to try to fix the issue with the bleeding around the edges of surfaces, like at the bottom of the pillars in the above examples.
Hm... But now that I take a second look at it, it looks like both versions have the same lightmapping issue.
And if I am not mistake, that second level is a maplet level?
So maybe pert or all of the issue is with how maplet sets up it's lightmapping coordinates.
But I dunno about that, cause there's no way to make lightmap coordinates line up perfectly with all the texels. So all games must have to make corrections for this issue during the lighting calculations themselves.
I wonder if one could solve this by calcualting a lightmap with 2-4x the detail level, and them averaging it down. That then would make it so that if a texel is mostly outside of an object, 4x as many pixels might be lit as unlit, and the averaging of them down to a single pixel color would result in a drastically reduced darkening effect on those pixels where the center of the texel is inside an object.
But I'm not sure that would entirely solve the issue. Maybe just reduce it by half.
I think what you might need to do is sample the 4 corners of each texel, and if any of those is exposed to the light, then the pixel is calculated as if it is in light.
Or you might be able to calcualte the lightmap at 2x it's width/height, size, and then take the maximum pixel color in each 4 pixel square when downsampling to a single pixel, instead of averaging them.