diff --git a/engine/shaders/glsl/defaultskin.glsl b/engine/shaders/glsl/defaultskin.glsl
index 1f72c445c..b8b5b89c5 100644
--- a/engine/shaders/glsl/defaultskin.glsl
+++ b/engine/shaders/glsl/defaultskin.glsl
@@ -72,7 +72,9 @@ void main ()
 	float d = dot(n,e_light_dir);
 	if (d < 0.0)		//vertex shader. this might get ugly, but I don't really want to make it per vertex.
 		d = 0.0;	//this avoids the dark side going below the ambient level.
-	light = e_light_ambient + (dot(n,e_light_dir)*e_light_mul);
+	light = e_light_ambient + (d*e_light_mul);
+
+//FIXME: Software rendering imitation should possibly push out normals by half a pixel or something to approximate software's over-estimation of distant model sizes (small models are drawn using JUST their verticies using the nearest pixel, which results in larger meshes)
 
 #ifdef TESS
 	normal = n;
@@ -236,7 +238,7 @@ void main ()
 	//FIXME: with this extra flag, half the permutations are redundant.
 	lightlev *= 0.5;	//counter the fact that the colourmap contains overbright values and logically ranges from 0 to 2 intead of to 1.
 	float pal = texture2D(s_paletted, tc).r;	//the palette index. hopefully not interpolated.
-	lightlev -= 1.0 / 128.0;	//software rendering appears to round down, so make sure we favour the lower values instead of rounding to the nearest
+//	lightlev -= 1.0 / 128.0;	//software rendering appears to round down, so make sure we favour the lower values instead of rounding to the nearest
 	col.r = texture2D(s_colourmap, vec2(pal, 1.0-lightlev.r)).r;	//do 3 lookups. this is to cope with lit files, would be a waste to not support those.
 	col.g = texture2D(s_colourmap, vec2(pal, 1.0-lightlev.g)).g;	//its not very softwarey, but re-palettizing is ugly.
 	col.b = texture2D(s_colourmap, vec2(pal, 1.0-lightlev.b)).b;	//without lits, it should be identical.