diff --git a/rogueviz/snow.cpp b/rogueviz/snow.cpp
new file mode 100644
index 00000000..1b964443
--- /dev/null
+++ b/rogueviz/snow.cpp
@@ -0,0 +1,147 @@
+#include "../hyper.h"
+
+/** \brief Snowball visualization
+ *
+ *  This visualization puts small objects ('snowballs') randomly throughout the space. 
+ *  It provides a way to visualize the geometry without any tessellation.
+ *
+ *  Should work for tessellations where every tile is congruent.
+ *
+ *  The snow_lambda parameter gives the expected number of snowballs per cell.
+ *  (The number in every region has Poisson distribution with mean proportional to its area.)
+ *
+ *  Not implemented for: product
+ *
+ **/
+
+namespace hr {
+
+ld snow_lambda = 1;
+
+bool snow_test = false;
+
+/* a funny glitch */
+bool snow_glitch = false;
+
+int snow_shape = 0;
+
+map<cell*, vector<transmatrix> > matrices_at;
+
+hpcshape& shapeid(int i) {
+  switch(i) {
+    case 0: 
+      return cgi.shSnowball;
+    case 1:
+      return cgi.shHeptaMarker;
+    case 2:
+      return cgi.shDisk;
+    default:
+      return cgi.shDisk;
+    }
+  }
+
+bool draw_snow(cell *c, const transmatrix& V) {
+  
+  if(!matrices_at.count(c)) {
+    auto& v = matrices_at[c];
+    int cnt = 0;
+    ld prob = randd();
+    ld poisson = exp(-snow_lambda);
+    while(cnt < 2*snow_lambda+100) {
+      if(prob < poisson) break;
+      prob -= poisson;
+      cnt++;
+      poisson *= snow_lambda / cnt;
+      }
+    if(snow_test) {
+      if(c != cwt.at) 
+        cnt = 0;
+      else {
+        c->wall = waFloorA;
+        cnt = snow_lambda;
+        }
+      }      
+    
+    if(snow_glitch) {
+      // in the standard tiling, this is incorrect but fun
+      for(int t=0; t<cnt; t++) {
+        hyperpoint h = C0;
+        h[0] = randd() - .5;
+        h[1] = randd() - .5;
+        h[2] = randd() - .5;
+        h[2] = -h[2];
+        v.push_back(rgpushxto0(h));
+        }
+      }
+    else if(nonisotropic || bt::in()) {
+
+      int co = bt::expansion_coordinate();
+      ld aer = bt::area_expansion_rate();
+
+      for(int t=0; t<cnt; t++) {
+        hyperpoint h;
+        // randd() - .5;
+        
+        for(int a=0; a<3; a++) {
+          if(a != co || aer == 1)
+            h[a] = randd() * 2 - 1;
+          else {
+            ld r = randd();
+            h[co] = log(lerp(1, aer, r)) / log(aer) * 2 - 1;
+            }
+          }
+        h = tC0(bt::normalized_at(h));
+        v.push_back(rgpushxto0(h));
+        }
+      }
+    else {
+      while(isize(v) < cnt) {
+        ld maxr = WDIM == 2 ? cgi.rhexf : cgi.corner_bonus;
+        ld vol = randd() * wvolarea_auto(maxr);
+        ld r = binsearch(0, maxr, [vol] (ld r) { return wvolarea_auto(r) > vol; });
+        transmatrix T = random_spin();
+        hyperpoint h = T * xpush0(r);
+        cell* c1 = c;
+        virtualRebase(c1, h);
+        if(c1 == c)
+          v.push_back(T * xpush(r));
+        }
+      }
+    }
+  
+  poly_outline = 0xFF;
+  for(auto& T: matrices_at[c])
+    queuepoly(V * T, shapeid(snow_shape), 0xFFFFFFFF).tinf = nullptr;
+
+  return false;
+  }
+
+bool cylanim = false;
+
+auto hchook = addHook(hooks_drawcell, 100, draw_snow)
+
++ addHook(clearmemory, 40, [] () {
+    matrices_at.clear();
+    })
+
++ addHook(hooks_args, 100, [] {
+  using namespace arg;
+           
+  if(0) ;
+  else if(argis("-snow-lambda")) {
+    shift_arg_formula(snow_lambda);
+    }
+  else if(argis("-snow-shape")) {
+    shift(); snow_shape = argi();
+    }
+  else if(argis("-snow-test")) {
+    snow_test = true;
+    }
+  else if(argis("-snow-glitch")) {
+    snow_test = true;
+    }
+  else return 1;
+  return 0;
+  });
+
+}