diff --git a/packages/ludic.render3d/env.ludic b/packages/ludic.render3d/env.ludic
index 1f8a9559..fc07c7f6 100644
--- a/packages/ludic.render3d/env.ludic
+++ b/packages/ludic.render3d/env.ludic
@@ -214,6 +214,7 @@ export state Render3dState {
   gvk_cp_layout: []long = null
   gvk_cp_dsl: []long = null
   gvk_cp_nbuf: []int = null
+  gvk_cp_ntex: []int = null
   gvk_pipe_keys: []string = null
   gvk_pipe: []long = null
   gvk_zero_vbuf: int = 0
@@ -363,8 +364,22 @@ export state Render3dState {
   grass_n: int = 0
   grass_cmds: int = 0
   grass_band_start: words = null    # per band: its first record
-  grass_band_cells: words = null    # ... and its 16 m cells per tile side
+  grass_band_cells: words = null    # ... and its cells per tile side
   grass_band_n: int = 0
+  gg_on: bool = false                # the blades are culled on the GPU (grass_gpu.ludic)
+  gg_reset: int = 0
+  gg_cull: int = 0
+  gg_prog: int = 0
+  gg_tiles_buf: []int = null
+  gg_tv: floats = null
+  gg_out: int = 0
+  gg_cmds: int = 0
+  gg_mesh: []Mesh = null
+  gg_n: int = 0
+  gg_max: int = 0
+  grass_band_mesh: []Mesh = null     # per band: the blade it draws (rows fall with distance)
+  grass_mesh3: Mesh = null           # three rows: the middle distance
+  grass_mesh2: Mesh = null           # two rows, one quad: far enough that the arch is under a pixel
   grass_mesh_on: bool = false
   grass_mesh_prog: int = 0
   r3d_draw_cb: fn() = null
@@ -590,6 +605,7 @@ export state Render3dState {
   sc_skip_blade: bool = false
   sc_skip_grass: bool = false
   grass_force: bool = false        # R3D_BLADES: the blades on whatever the game set
+  grass_env_off: bool = false      # R3D_NOBLADES: the blades off whatever the game set
   sc_skip_card: bool = false
   cb_state: int = 12345
   shadow_res: int = 2048
diff --git a/packages/ludic.render3d/gpu.ludic b/packages/ludic.render3d/gpu.ludic
index 90489d07..49fea526 100644
--- a/packages/ludic.render3d/gpu.ludic
+++ b/packages/ludic.render3d/gpu.ludic
@@ -513,6 +513,14 @@ function gpu_compute(render3d_st: mut Render3dState, name: string, n_bufs: int)
 function gpu_dispatch(render3d_st: mut Render3dState, c: int, params: pointer, n_params: int, bufs: words, groups: int) -> void {
   if render3d_st.gpu_kind == GPU_VK and c > 0 { gvk_dispatch(render3d_st, c, params, n_params, bufs, groups, 1, 1) }
 }
+# a compute program that also samples n_tex textures (bound after its buffers), and its 2-D dispatch
+function gpu_compute_tex(render3d_st: mut Render3dState, name: string, n_bufs: int, n_tex: int) -> int {
+  if render3d_st.gpu_kind != GPU_VK { return 0 }
+  return gvk_compute_new_tex(render3d_st, name, n_bufs, n_tex)
+}
+function gpu_dispatch_tex(render3d_st: mut Render3dState, c: int, params: pointer, n_params: int, bufs: words, texs: words, gx: int, gy: int) -> void {
+  if render3d_st.gpu_kind == GPU_VK and c > 0 { gvk_dispatch_tex(render3d_st, c, params, n_params, bufs, texs, gx, gy, 1) }
+}
 # a buffer a compute pass writes (never reallocated under a draw that reads it)
 function gpu_buffer_gpu_owned(render3d_st: mut Render3dState, buf: int) -> void { if render3d_st.gpu_kind == GPU_VK { gvk_buf_gpu_owned(render3d_st, buf) } }
 # the host-visible contents of a buffer, for a readback after gpu_finish; null on OpenGL
diff --git a/packages/ludic.render3d/gpu_vk_draw.ludic b/packages/ludic.render3d/gpu_vk_draw.ludic
index 967ee298..b5bb3acb 100644
--- a/packages/ludic.render3d/gpu_vk_draw.ludic
+++ b/packages/ludic.render3d/gpu_vk_draw.ludic
@@ -120,24 +120,28 @@ function gvk_program(render3d_st: mut Render3dState, p: int, key: string, spv_di
 
 # ---- compute ------------------------------------------------------------------------------
 # A compute program is <name>.comp.spv beside the manifest, with bindings fixed by convention:
-# 0 the parameter block (a uniform buffer, copied into the frame's ring at the dispatch) and
-# 1 .. n storage buffers. Handles start at 1.
+# 0 the parameter block (a uniform buffer, copied into the frame's ring at the dispatch),
+# 1 .. n storage buffers, then n+1 .. n+m sampled textures (linear, clamped). Handles start at 1.
 
-function gvk_compute_new(render3d_st: mut Render3dState, name: string, n_bufs: int) -> int {
+function gvk_compute_new(render3d_st: mut Render3dState, name: string, n_bufs: int) -> int { return gvk_compute_new_tex(render3d_st, name, n_bufs, 0) }
+
+function gvk_compute_new_tex(render3d_st: mut Render3dState, name: string, n_bufs: int, n_tex: int) -> int {
   let zero: long = 0
   if render3d_st.gvk_cp_pipe == null {
     render3d_st.gvk_cp_pipe = new []long;  render3d_st.gvk_cp_layout = new []long;  render3d_st.gvk_cp_dsl = new []long;  render3d_st.gvk_cp_nbuf = new []int
-    push(render3d_st.gvk_cp_pipe, zero);  push(render3d_st.gvk_cp_layout, zero);  push(render3d_st.gvk_cp_dsl, zero);  push(render3d_st.gvk_cp_nbuf, 0)
+    render3d_st.gvk_cp_ntex = new []int
+    push(render3d_st.gvk_cp_pipe, zero);  push(render3d_st.gvk_cp_layout, zero);  push(render3d_st.gvk_cp_dsl, zero);  push(render3d_st.gvk_cp_nbuf, 0);  push(render3d_st.gvk_cp_ntex, 0)
   }
   let module = gvk_module(render3d_st, `{render3d_st.gvk_spv_dir}/{name}.comp.spv`)
   if module == 0 { return 0 }
-  let nb = n_bufs + 1
+  let nb = n_bufs + 1 + n_tex
   let bw = VkDescriptorSetLayoutBinding_sizeof
   let binds = bytes(bw * nb)
   Vk.zero(binds, bw * nb)
   for k in 0 .. nb {
     var kind = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER
     if k == 0 { kind = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER }
+    if k > n_bufs { kind = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER }
     Vk.put_i32(binds, k * bw + VkDescriptorSetLayoutBinding_binding, k)
     Vk.put_i32(binds, k * bw + VkDescriptorSetLayoutBinding_descriptorType, kind)
     Vk.put_i32(binds, k * bw + VkDescriptorSetLayoutBinding_descriptorCount, 1)
@@ -172,17 +176,25 @@ function gvk_compute_new(render3d_st: mut Render3dState, name: string, n_bufs: i
   r = Vk.create_compute_pipelines(render3d_st.gvk_dev, zero, 1, cpci, null, out)
   if r != VK_SUCCESS { gvk_fail(render3d_st, `vkCreateComputePipelines {name}`, r);  return 0 }
   push(render3d_st.gvk_cp_pipe, gvk_handle(out));  push(render3d_st.gvk_cp_layout, layout);  push(render3d_st.gvk_cp_dsl, Vk.get_i64(dsl, 0));  push(render3d_st.gvk_cp_nbuf, n_bufs)
+  push(render3d_st.gvk_cp_ntex, n_tex)
   return len(render3d_st.gvk_cp_pipe) - 1
 }
 
 # Record a dispatch of compute program c into the frame, between passes: params (n_params
 # bytes, std140) as binding 0 and bufs[0 .. n) as bindings 1 .. n.
-function gvk_dispatch(render3d_st: mut Render3dState, c: int, params: pointer, n_params: int, bufs: words, gx: int, gy: int, gz: int) -> void {
+function gvk_dispatch(render3d_st: mut Render3dState, c: int, params: pointer, n_params: int, bufs: words, gx: int, gy: int, gz: int) -> void { gvk_dispatch_tex(render3d_st, c, params, n_params, bufs, null, gx, gy, gz) }
+
+# ... and texs[0 .. m) as the sampled textures after the buffers. What it writes is made visible to
+# the indirect draws, vertex attributes and shaders that follow (Metal's own hazard tracking was all
+# that ordered the tree culling's output before this)
+function gvk_dispatch_tex(render3d_st: mut Render3dState, c: int, params: pointer, n_params: int, bufs: words, texs: words, gx: int, gy: int, gz: int) -> void {
   let zero: long = 0
   if render3d_st.gvk_cp_pipe == null or c <= 0 or c >= len(render3d_st.gvk_cp_pipe) { return }
   gvk_pass_end(render3d_st)
   let cb = gvk_frame_cb(render3d_st)
   let nb = render3d_st.gvk_cp_nbuf[c]
+  var nt = 0
+  if render3d_st.gvk_cp_ntex != null and c < len(render3d_st.gvk_cp_ntex) { nt = render3d_st.gvk_cp_ntex[c] }
   let dsai = bytes(VkDescriptorSetAllocateInfo_sizeof)
   Vk.zero(dsai, VkDescriptorSetAllocateInfo_sizeof)
   Vk.put_i32(dsai, VkDescriptorSetAllocateInfo_sType, VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO)
@@ -199,10 +211,13 @@ function gvk_dispatch(render3d_st: mut Render3dState, c: int, params: pointer, n
   if at < 0 { print("r3d: vulkan: the frame's uniform ring is full");  return }
   let ww = VkWriteDescriptorSet_sizeof
   let bw = VkDescriptorBufferInfo_sizeof
-  let writes = bytes(ww * (nb + 1))
-  Vk.zero(writes, ww * (nb + 1))
+  let iw = VkDescriptorImageInfo_sizeof
+  let writes = bytes(ww * (nb + 1 + nt))
+  Vk.zero(writes, ww * (nb + 1 + nt))
   let infos = bytes(bw * (nb + 1))
   Vk.zero(infos, bw * (nb + 1))
+  let iis = bytes(iw * (nt + 1))
+  Vk.zero(iis, iw * (nt + 1))
   for k in 0 .. nb + 1 {
     var buf = render3d_st.gvk_ring_buf
     var off: long = 0
@@ -220,10 +235,31 @@ function gvk_dispatch(render3d_st: mut Render3dState, c: int, params: pointer, n
     Vk.put_i32(writes, k * ww + VkWriteDescriptorSet_descriptorType, kind)
     Vk.put_ptr(writes, k * ww + VkWriteDescriptorSet_pBufferInfo, mem_off(infos, k * bw))
   }
-  Vk.update_descriptor_sets(render3d_st.gvk_dev, nb + 1, writes, 0, null)
+  let smp = gvk_sampler(render3d_st, GL_LINEAR, GL_LINEAR, GL_CLAMP_TO_EDGE, GL_CLAMP_TO_EDGE, 0, 0)
+  for t in 0 .. nt {
+    var tex = render3d_st.gvk_white
+    if texs != null and t < len(texs) and texs[t] > 0 { tex = texs[t] }
+    Vk.put_i64(iis, t * iw + VkDescriptorImageInfo_sampler, smp)
+    Vk.put_i64(iis, t * iw + VkDescriptorImageInfo_imageView, render3d_st.gvk_tex_view[tex])
+    Vk.put_i32(iis, t * iw + VkDescriptorImageInfo_imageLayout, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL)
+    let k = nb + 1 + t
+    Vk.put_i32(writes, k * ww + VkWriteDescriptorSet_sType, VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET)
+    Vk.put_i64(writes, k * ww + VkWriteDescriptorSet_dstSet, set)
+    Vk.put_i32(writes, k * ww + VkWriteDescriptorSet_dstBinding, k)
+    Vk.put_i32(writes, k * ww + VkWriteDescriptorSet_descriptorCount, 1)
+    Vk.put_i32(writes, k * ww + VkWriteDescriptorSet_descriptorType, VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER)
+    Vk.put_ptr(writes, k * ww + VkWriteDescriptorSet_pImageInfo, mem_off(iis, t * iw))
+  }
+  Vk.update_descriptor_sets(render3d_st.gvk_dev, nb + 1 + nt, writes, 0, null)
   Vk.cmd_bind_pipeline(cb, VK_PIPELINE_BIND_POINT_COMPUTE, render3d_st.gvk_cp_pipe[c])
   Vk.cmd_bind_descriptor_sets(cb, VK_PIPELINE_BIND_POINT_COMPUTE, render3d_st.gvk_cp_layout[c], 0, 1, sets, 0, null)
   Vk.cmd_dispatch(cb, gx, gy, gz)
+  let mb = bytes(VkMemoryBarrier_sizeof)
+  Vk.zero(mb, VkMemoryBarrier_sizeof)
+  Vk.put_i32(mb, VkMemoryBarrier_sType, VK_STRUCTURE_TYPE_MEMORY_BARRIER)
+  Vk.put_i32(mb, VkMemoryBarrier_srcAccessMask, VK_ACCESS_SHADER_WRITE_BIT)
+  Vk.put_i32(mb, VkMemoryBarrier_dstAccessMask, VK_ACCESS_INDIRECT_COMMAND_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT)
+  Vk.cmd_pipeline_barrier(cb, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT | VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 1, mb, 0, null, 0, null)
 }
 
 # R3D_VK_PROBE=1: one dispatch over a GPU-owned buffer, read back - the compute path end to end
diff --git a/packages/ludic.render3d/grass.ludic b/packages/ludic.render3d/grass.ludic
index 88152c55..27641155 100644
--- a/packages/ludic.render3d/grass.ludic
+++ b/packages/ludic.render3d/grass.ludic
@@ -1,7 +1,7 @@
 # ============================================================================
 # grass.ludic — procedural GPU ground cover with continuous density (no rings).
 #
-# The world is cut into 16 m cells; blade j of a cell always stands in the same place
+# The world is cut into 4 m cells; blade j of a cell always stands in the same place
 # (shaders/grass.vert). Draws are per tile: the CPU walks tiles around the camera,
 # frustum-culls them, and feeds each visible tile as many blade indices per cell as its
 # NEAREST point could need; the vertex stage then keeps only the indices that exist at
@@ -10,7 +10,7 @@
 # draw count down — the cells and their hashes are the same in every tile size.
 # ============================================================================
 
-const GRASS_CELL: int = 16
+const GRASS_CELL: int = 4
 # A photographed blade, as an atlas of straightened blades side by side (the game sets
 # this; the renderer does not name a game asset). 0 = the procedural gradient, which is
 # what this was for a year: a two-tone ramp with a hard edge, and every blade in the
@@ -89,11 +89,13 @@ function grass_blade_mesh(render3d_st: mut Render3dState, rows: int) -> Mesh {
 
 function grass_init(render3d_st: mut Render3dState) -> void {
   render3d_st.grass_merge = gpu_has_mdi(render3d_st) and not r3d_env_has(render3d_st, "R3D_GRASS_TILES")
-  var defs = "#define FOLIAGE\n#define BLADE\n"
+  var defs = "#define FOLIAGE\n#define BLADE\n#define GBLADE\n"
   if render3d_st.grass_merge {
     defs = defs + "#define TILES\n"
     render3d_st.grass_rec = words(GRASS_RECS * 5);  render3d_st.grass_tv = floats(GRASS_RECS * 4);  render3d_st.grass_chunk_tv = floats(GRASS_CHUNK * 4)
     render3d_st.grass_band_start = words(4);  render3d_st.grass_band_cells = words(4)
+    render3d_st.grass_band_mesh = new []Mesh
+    for b in 0 .. 4 { push(render3d_st.grass_band_mesh, null) }
     render3d_st.grass_cmds = gpu_buffer_new(render3d_st)
   }
   render3d_st.grass_prog = r3d_program(render3d_st, "grass.vert", "model.frag", defs)
@@ -101,6 +103,8 @@ function grass_init(render3d_st: mut Render3dState) -> void {
   # five rows, four quads: the arch above needs somewhere to bend, and at four rows a
   # blade that leans over is three straight segments and shows every join
   render3d_st.grass_mesh = grass_blade_mesh(render3d_st, 5)
+  render3d_st.grass_mesh3 = grass_blade_mesh(render3d_st, 3)
+  render3d_st.grass_mesh2 = grass_blade_mesh(render3d_st, 2)
   render3d_st.grass_wind = 2.4
   # Matched to the blade's width: a 1 cm blade at 0.11 m spacing covers a third of what a
   # 2.8 cm blade did, and the meadow goes bare. The game's graphics settings override this
@@ -108,14 +112,17 @@ function grass_init(render3d_st: mut Render3dState) -> void {
   # gets there, so the two have to agree or a shot shows something no player will see.
   # That is exactly how the last change measured as "no effect": the render was identical
   # because this line, not the settings, was deciding.
+  # The spacing doubles every 18 m and the blades stop at 70 m: at 45 m and 1600 m a 1 cm blade
+  # was under a pixel from 20 m on, costing a full blade's vertices to shimmer (5 ms of 9.7 on GL).
   render3d_st.grass_s0 = 0.066
-  render3d_st.grass_d0 = 45.0
-  render3d_st.grass_radius = 1600.0
+  render3d_st.grass_d0 = 18.0
+  render3d_st.grass_radius = 70.0
   if r3d_env_has(render3d_st, "R3D_NOBLADES") { render3d_st.grass_on = false }
   if r3d_env_has(render3d_st, "R3D_GRASS_R") { render3d_st.grass_radius = float(Text.to_int(r3d_env(render3d_st, "R3D_GRASS_R"))) }
   if r3d_env_has(render3d_st, "R3D_GRASS_S0") { render3d_st.grass_s0 = float(Text.to_int(r3d_env(render3d_st, "R3D_GRASS_S0"))) / 1000.0 }
   if r3d_env_has(render3d_st, "R3D_GRASS_D0") { render3d_st.grass_d0 = float(Text.to_int(r3d_env(render3d_st, "R3D_GRASS_D0"))) }
   if r3d_env_has(render3d_st, "R3D_GRASS_DBG") { render3d_st.grass_dbg = Text.to_int(r3d_env(render3d_st, "R3D_GRASS_DBG")) }
+  gg_init(render3d_st)
 }
 
 # indices per 16 m cell that could exist at distance d (the count the shader computes)
@@ -126,11 +133,13 @@ function grass_count_at(render3d_st: Render3dState, d: float) -> int {
 }
 
 # one tile size over one distance band
-function grass_tiles(render3d_st: mut Render3dState, size: int, d_min: float, d_max: float) -> void {
+function grass_tiles(render3d_st: mut Render3dState, size: int, d_min: float, d_max: float, blade: Mesh) -> void {
   let p = render3d_st.grass_prog
   let cells = size / GRASS_CELL
+  if d_min >= render3d_st.grass_radius { return }
   if render3d_st.grass_merge {
-    render3d_st.grass_band_start[render3d_st.grass_band_n] = render3d_st.grass_n;  render3d_st.grass_band_cells[render3d_st.grass_band_n] = cells;  render3d_st.grass_band_n += 1
+    render3d_st.grass_band_start[render3d_st.grass_band_n] = render3d_st.grass_n;  render3d_st.grass_band_cells[render3d_st.grass_band_n] = cells
+    render3d_st.grass_band_mesh[render3d_st.grass_band_n] = blade;  render3d_st.grass_band_n += 1
   } else {
     u_i(render3d_st, gpu_uniform(render3d_st, p, "u_tile_cells"), cells)
   }
@@ -161,7 +170,7 @@ function grass_tiles(render3d_st: mut Render3dState, size: int, d_min: float, d_
             var inst = per * cells * cells
             if inst > 65535 { inst = 65535 }
             let r = render3d_st.grass_n * 5
-            render3d_st.grass_rec[r] = render3d_st.grass_mesh.count;  render3d_st.grass_rec[r + 1] = inst;  render3d_st.grass_rec[r + 2] = 0;  render3d_st.grass_rec[r + 3] = 0
+            render3d_st.grass_rec[r] = blade.count;  render3d_st.grass_rec[r + 1] = inst;  render3d_st.grass_rec[r + 2] = 0;  render3d_st.grass_rec[r + 3] = 0
             render3d_st.grass_rec[r + 4] = ((render3d_st.grass_n - render3d_st.grass_band_start[render3d_st.grass_band_n - 1]) % GRASS_CHUNK) * 65536
             let t = render3d_st.grass_n * 4
             render3d_st.grass_tv[t] = ox;  render3d_st.grass_tv[t + 1] = oz;  render3d_st.grass_tv[t + 2] = float(per);  render3d_st.grass_tv[t + 3] = 0.0
@@ -170,7 +179,7 @@ function grass_tiles(render3d_st: mut Render3dState, size: int, d_min: float, d_
         } else if per > 0 {
           u_f2(render3d_st, gpu_uniform(render3d_st, p, "u_tile"), ox, oz)
           u_i(render3d_st, gpu_uniform(render3d_st, p, "u_per_cell"), per)
-          mesh_draw_instanced(render3d_st, render3d_st.grass_mesh, per * cells * cells)
+          mesh_draw_instanced(render3d_st, blade, per * cells * cells)
           render3d_st.grass_draws += 1
         }
       }
@@ -180,11 +189,27 @@ function grass_tiles(render3d_st: mut Render3dState, size: int, d_min: float, d_
   }
 }
 
+function grass_live(render3d_st: Render3dState) -> bool { return not render3d_st.grass_env_off and (render3d_st.grass_on or render3d_st.grass_force) and render3d_st.grass_prog != 0 }
+
 function grass_draw(render3d_st: mut Render3dState) -> void {
-  if not (render3d_st.grass_on or render3d_st.grass_force) or render3d_st.ter_reflect or render3d_st.grass_prog == 0 { return }
+  if not grass_live(render3d_st) or render3d_st.ter_reflect { return }
+  if render3d_st.gg_on { gg_draw(render3d_st);  return }
   var p = render3d_st.grass_prog
   if grass_mesh_live(render3d_st) { p = render3d_st.grass_mesh_prog }
   gpu_use_program(render3d_st, p)
+  grass_bind(render3d_st, p)
+  gpu_cull(render3d_st, false)
+  render3d_st.grass_draws = 0
+  render3d_st.grass_n = 0
+  render3d_st.grass_band_n = 0
+  gpu_mesh_bind(render3d_st, render3d_st.grass_mesh)
+  grass_bands(render3d_st)
+  if render3d_st.grass_merge { grass_flush(render3d_st) }
+  gpu_cull(render3d_st, true)
+}
+
+# the blades' uniforms, shared by the per-vertex path and the culled one (grass_gpu.ludic)
+function grass_bind(render3d_st: mut Render3dState, p: int) -> void {
   u_mat4(render3d_st, gpu_uniform(render3d_st, p, "u_view"), render3d_st.cam_view)
   u_mat4(render3d_st, gpu_uniform(render3d_st, p, "u_proj"), render3d_st.cam_proj)
   u_mat4(render3d_st, gpu_uniform(render3d_st, p, "u_vp"), render3d_st.cam_vp_clean)
@@ -207,6 +232,9 @@ function grass_draw(render3d_st: mut Render3dState) -> void {
   u_f(render3d_st, gpu_uniform(render3d_st, p, "u_s0"), render3d_st.grass_s0)
   u_f(render3d_st, gpu_uniform(render3d_st, p, "u_d0"), render3d_st.grass_d0)
   u_f(render3d_st, gpu_uniform(render3d_st, p, "u_radius"), render3d_st.grass_radius)
+  var ph = render3d_st.post_h
+  if ph <= 0 { ph = gl_height() }
+  u_f(render3d_st, gpu_uniform(render3d_st, p, "u_px"), 2.0 * Math.tan(render3d_st.cam_fov * 0.5) / float(ph))
   u_i(render3d_st, gpu_uniform(render3d_st, p, "u_dbg"), render3d_st.grass_dbg)
   var orthotex = render3d_st.ter_ortho_tex
   var oon = 1.0
@@ -230,16 +258,16 @@ function grass_draw(render3d_st: mut Render3dState) -> void {
   # does have a faint sheen, so this is small rather than nothing - the foliage layers have
   # used 0.05 all along and never showed the fault.
   u_f(render3d_st, gpu_uniform(render3d_st, p, "u_spec_scale"), 0.008)
-  gpu_cull(render3d_st, false)
-  render3d_st.grass_draws = 0
-  render3d_st.grass_n = 0
-  render3d_st.grass_band_n = 0
-  gpu_mesh_bind(render3d_st, render3d_st.grass_mesh)
-  grass_tiles(render3d_st, 16, 0.0, 300.0)
-  grass_tiles(render3d_st, 64, 300.0, 1200.0)
-  grass_tiles(render3d_st, 256, 1200.0, render3d_st.grass_radius)
-  if render3d_st.grass_merge { grass_flush(render3d_st) }
-  gpu_cull(render3d_st, true)
+}
+
+# the per-vertex path's tiles, in four bands
+function grass_bands(render3d_st: mut Render3dState) -> void {
+  # four bands: the tile grows and the blade loses rows with distance (a band past the radius is
+  # skipped); a tile asks for what its NEAREST point needs, so a wide tile far out wastes the rest
+  grass_tiles(render3d_st, 4, 0.0, 6.0, render3d_st.grass_mesh)
+  grass_tiles(render3d_st, 4, 6.0, 16.0, render3d_st.grass_mesh3)
+  grass_tiles(render3d_st, 8, 16.0, 40.0, render3d_st.grass_mesh2)
+  grass_tiles(render3d_st, 16, 40.0, render3d_st.grass_radius, render3d_st.grass_mesh2)
 }
 
 # the records gathered this frame: uploaded once, then each band GRASS_CHUNK records a draw
@@ -253,6 +281,8 @@ function grass_flush(render3d_st: mut Render3dState) -> void {
     var e = render3d_st.grass_n
     if b + 1 < render3d_st.grass_band_n { e = render3d_st.grass_band_start[b + 1] }
     if e > s { u_i(render3d_st, gpu_uniform(render3d_st, p, "u_tile_cells"), render3d_st.grass_band_cells[b]) }
+    let blade = render3d_st.grass_band_mesh[b]
+    if e > s and not mesh { gpu_mesh_bind(render3d_st, blade) }
     var k = s
     while k < e {
       var m = e - k
@@ -275,7 +305,7 @@ function grass_flush(render3d_st: mut Render3dState) -> void {
         }
       } else {
         u_f4v(render3d_st, gpu_uniform(render3d_st, p, "u_tiles"), m, render3d_st.grass_chunk_tv)
-        gpu_draw_mesh_indirect(render3d_st, render3d_st.grass_mesh, render3d_st.grass_cmds, k * 20, m, 0, 0)
+        gpu_draw_mesh_indirect(render3d_st, blade, render3d_st.grass_cmds, k * 20, m, 0, 0)
       }
       if not mesh { render3d_st.grass_draws += 1 }
       k += m
diff --git a/packages/ludic.render3d/grass_gpu.ludic b/packages/ludic.render3d/grass_gpu.ludic
new file mode 100644
index 00000000..845ea613
--- /dev/null
+++ b/packages/ludic.render3d/grass_gpu.ludic
@@ -0,0 +1,154 @@
+# ============================================================================
+# grass_gpu.ludic — the meadow's blades culled on the GPU (Vulkan), drawn from what survived.
+#
+# grass.vert decided every blade per VERTEX: its hashes, the ground under it, whether anything
+# grows there, its colour field - ten times for a five-row blade, and in full for every index a
+# tile asked for and threw away. That was 3.5 ms of a MoltenVK frame. Here the CPU still walks the
+# visible tiles (as grass_tiles does), but grass_cull.comp decides each candidate once and writes
+# the survivors into three bands by distance, one indirect draw each; grass_inst.vert only bends
+# and places the vertices. OpenGL keeps the per-vertex path. R3D_GRASS_GPU=0 does too, to compare.
+# ============================================================================
+
+const GG_TILES: int = 4096
+const GG_BANDS: int = 3
+const GG_NEAR: float = 6.0     # band 0, five rows
+const GG_MID: float = 16.0     # band 1, three rows; band 2 is one quad
+
+function gg_cap(b: int) -> int {
+  if b == 0 { return 16384 }
+  if b == 1 { return 49152 }
+  return 163840
+}
+function gg_base(b: int) -> int {
+  if b == 0 { return 0 }
+  if b == 1 { return 16384 }
+  return 65536
+}
+
+# once, after grass_init: the programs, the buffers and the three instanced blades
+function gg_init(render3d_st: mut Render3dState) -> void {
+  if not gpu_has_compute(render3d_st) or render3d_st.grass_prog == 0 { return }
+  if r3d_env_has(render3d_st, "R3D_GRASS_GPU") and r3d_env(render3d_st, "R3D_GRASS_GPU") == "0" { return }
+  render3d_st.gg_reset = gpu_compute(render3d_st, "grass_reset", 1)
+  render3d_st.gg_cull = gpu_compute_tex(render3d_st, "grass_cull", 3, 2)
+  render3d_st.gg_prog = r3d_program(render3d_st, "grass_inst.vert", "model.frag", "#define FOLIAGE\n#define BLADE\n#define GBLADE\n#define GINST\n")
+  if render3d_st.gg_reset == 0 or render3d_st.gg_cull == 0 or render3d_st.gg_prog == 0 { return }
+  render3d_st.gg_tiles_buf = new []int
+  for k in 0 .. 2 { push(render3d_st.gg_tiles_buf, gpu_buffer_new(render3d_st)) }
+  render3d_st.gg_tv = floats(GG_TILES * 4)
+  render3d_st.gg_out = gpu_buffer_new(render3d_st)
+  gpu_buffer_upload(render3d_st, render3d_st.gg_out, (gg_base(2) + gg_cap(2)) * 64, null, GPU_DYNAMIC)
+  gpu_buffer_gpu_owned(render3d_st, render3d_st.gg_out)
+  render3d_st.gg_mesh = new []Mesh
+  push(render3d_st.gg_mesh, gg_blade(render3d_st, 5))
+  push(render3d_st.gg_mesh, gg_blade(render3d_st, 3))
+  push(render3d_st.gg_mesh, gg_blade(render3d_st, 2))
+  let rec = words(GG_BANDS * 5)
+  for b in 0 .. GG_BANDS {
+    rec[b * 5] = render3d_st.gg_mesh[b].count;  rec[b * 5 + 1] = 0;  rec[b * 5 + 2] = 0;  rec[b * 5 + 3] = 0;  rec[b * 5 + 4] = gg_base(b)
+  }
+  render3d_st.gg_cmds = gpu_buffer_new(render3d_st)
+  gpu_buffer_upload(render3d_st, render3d_st.gg_cmds, GG_BANDS * 20, data_of(rec), GPU_DYNAMIC)
+  gpu_buffer_gpu_owned(render3d_st, render3d_st.gg_cmds)
+  render3d_st.gg_on = true
+  print("r3d: grass: blades are culled on the GPU")
+}
+
+# a blade mesh reading its instance record (four vec4s) from the cull's output
+function gg_blade(render3d_st: mut Render3dState, rows: int) -> Mesh {
+  let m = grass_blade_mesh(render3d_st, rows)
+  gpu_mesh_bind_instances(render3d_st, m, render3d_st.gg_out)
+  for k in 0 .. 4 { gpu_mesh_attr_inst(render3d_st, m, 3 + k, 4, GPU_F32, 64, k * 16) }
+  gpu_mesh_done(render3d_st, m)
+  return m
+}
+
+# one tile size over one distance band: its visible tiles into gg_tv (corner, indices per cell, cells)
+function gg_tiles(render3d_st: mut Render3dState, size: int, d_min: float, d_max: float) -> void {
+  if d_min >= render3d_st.grass_radius { return }
+  let cells = size / GRASS_CELL
+  let sz = float(size)
+  let half = sz * 0.5
+  let reach = d_max + half * 1.5
+  let corner_r = half * 1.42
+  let tx0 = int(Math.floor((render3d_st.cam_pos[0] - reach) / sz))
+  let tx1 = int(Math.floor((render3d_st.cam_pos[0] + reach) / sz))
+  let tz0 = int(Math.floor((render3d_st.cam_pos[2] - reach) / sz))
+  let tz1 = int(Math.floor((render3d_st.cam_pos[2] + reach) / sz))
+  for tz in tz0 .. tz1 + 1 {
+    for tx in tx0 .. tx1 + 1 {
+      let ox = float(tx) * sz;  let oz = float(tz) * sz
+      let dx = ox + half - render3d_st.cam_pos[0];  let dz = oz + half - render3d_st.cam_pos[2]
+      let dc = Math.sqrt(dx * dx + dz * dz)
+      let dnear = Math.max(dc - corner_r, 0.0)
+      if dc < d_min or not (dnear < d_max) or render3d_st.gg_n >= GG_TILES { continue }
+      if not cam_sphere_visible(render3d_st, ox + half, terrain_height(render3d_st, ox + half, oz + half), oz + half, corner_r + 6.0) { continue }
+      let per = grass_count_at(render3d_st, dnear)
+      var total = per * cells * cells
+      if total > 65535 { total = 65535 }
+      let t = render3d_st.gg_n * 4
+      render3d_st.gg_tv[t] = ox;  render3d_st.gg_tv[t + 1] = oz;  render3d_st.gg_tv[t + 2] = float(total / (cells * cells));  render3d_st.gg_tv[t + 3] = float(cells)
+      if total > render3d_st.gg_max { render3d_st.gg_max = total }
+      render3d_st.gg_n += 1
+    }
+  }
+}
+
+# before the frame's first pass: the tiles, then the reset and the cull
+function gg_cull_frame(render3d_st: mut Render3dState) -> void {
+  if not render3d_st.gg_on or not grass_live(render3d_st) { return }
+  render3d_st.gg_n = 0
+  render3d_st.gg_max = 0
+  gg_tiles(render3d_st, 4, 0.0, 16.0)
+  gg_tiles(render3d_st, 8, 16.0, 40.0)
+  gg_tiles(render3d_st, 16, 40.0, render3d_st.grass_radius)
+  let rb = words(4)
+  rb[0] = GG_BANDS;  rb[1] = 0;  rb[2] = 0;  rb[3] = 0
+  let cb = words(1)
+  cb[0] = render3d_st.gg_cmds
+  gpu_dispatch(render3d_st, render3d_st.gg_reset, data_of(rb), 16, cb, 1)
+  if render3d_st.gg_n == 0 { return }
+  # the tile list alternates buffers by frame, so the frame still on the GPU keeps its own
+  let tb = render3d_st.gg_tiles_buf[render3d_st.r3d_test_frame % 2]
+  gpu_buffer_upload(render3d_st, tb, render3d_st.gg_n * 16, data_of(render3d_st.gg_tv), GPU_DYNAMIC)
+  let pr = gg_params(render3d_st)
+  let bufs = words(3)
+  bufs[0] = tb;  bufs[1] = render3d_st.gg_out;  bufs[2] = render3d_st.gg_cmds
+  let texs = words(2)
+  texs[0] = render3d_st.ter_height_tex;  texs[1] = render3d_st.ter_ortho_tex
+  gpu_dispatch_tex(render3d_st, render3d_st.gg_cull, data_of(pr), 192, bufs, texs, (render3d_st.gg_max + 63) / 64, render3d_st.gg_n)
+}
+
+# grass_cull.comp's Params, std140: 12 vec4s
+function gg_params(render3d_st: Render3dState) -> words {
+  let pr = words(48)
+  for i in 0 .. 48 { pr[i] = 0 }
+  if render3d_st.cam_planes != null { for i in 0 .. 16 { pr[i] = float_bits(render3d_st.cam_planes[i]) } }
+  pr[16] = float_bits(render3d_st.cam_pos[0]);  pr[17] = float_bits(render3d_st.cam_pos[1]);  pr[18] = float_bits(render3d_st.cam_pos[2]);  pr[19] = float_bits(render3d_st.grass_radius)
+  var ph = render3d_st.post_h
+  if ph <= 0 { ph = gl_height() }
+  pr[20] = float_bits(render3d_st.grass_s0);  pr[21] = float_bits(render3d_st.grass_d0);  pr[22] = float_bits(2.0 * Math.tan(render3d_st.cam_fov * 0.5) / float(ph));  pr[23] = float_bits(render3d_st.ter_snow_line)
+  pr[24] = float_bits(render3d_st.ter_lake_cx);  pr[25] = float_bits(render3d_st.ter_lake_cz);  pr[26] = float_bits(render3d_st.ter_lake_ex);  pr[27] = float_bits(render3d_st.ter_lake_ez)
+  var lake = -100000.0
+  if render3d_st.ter_lake_ex != 0.0 { lake = render3d_st.ter_lake_level }
+  var sea = lake
+  if render3d_st.ter_sea_set { sea = render3d_st.ter_sea_level }
+  var oon = 0.0
+  if render3d_st.ter_ortho_tex != 0 { oon = 1.0 }
+  pr[28] = float_bits(lake);  pr[29] = float_bits(sea);  pr[30] = float_bits(oon)
+  pr[32] = float_bits(render3d_st.ter_ox);  pr[33] = float_bits(render3d_st.ter_oz);  pr[34] = float_bits(float(render3d_st.TERRAIN_HALF))
+  pr[36] = float_bits(GG_NEAR);  pr[37] = float_bits(GG_MID)
+  for b in 0 .. GG_BANDS { pr[40 + b] = gg_base(b);  pr[44 + b] = gg_cap(b) }
+  return pr
+}
+
+# the survivors: one indirect draw per band, from the records grass_cull.comp counted into
+function gg_draw(render3d_st: mut Render3dState) -> void {
+  let p = render3d_st.gg_prog
+  gpu_use_program(render3d_st, p)
+  grass_bind(render3d_st, p)
+  u_f(render3d_st, gpu_uniform(render3d_st, p, "u_time"), render3d_st.r3d_time)
+  gpu_cull(render3d_st, false)
+  for b in 0 .. GG_BANDS { gpu_draw_mesh_indirect(render3d_st, render3d_st.gg_mesh[b], render3d_st.gg_cmds, b * 20, 1, 0, 0) }
+  gpu_cull(render3d_st, true)
+}
diff --git a/packages/ludic.render3d/programs.ludic b/packages/ludic.render3d/programs.ludic
index c3140758..ac82e165 100644
--- a/packages/ludic.render3d/programs.ludic
+++ b/packages/ludic.render3d/programs.ludic
@@ -51,6 +51,8 @@ function r3d_shader_src(render3d_st: mut Render3dState, name: string, defines: s
   if render3d_st.r3d_wind_src == null { render3d_st.r3d_wind_src = r3d_shader_file(render3d_st, "wind.glsl") }
   var s = "#version 410 core\n" + render3d_st.r3d_global_defs + defines + render3d_st.r3d_wind_src
   if is_frag { s = s + render3d_st.r3d_noise_src + render3d_st.r3d_lighting_src }
+  # a blade's colour field is worked out once per vertex, not once per pixel (grass.vert)
+  else if Text.contains(defines, "#define GBLADE") { s = s + render3d_st.r3d_noise_src }
   return s + r3d_shader_file(render3d_st, name)
 }
 
diff --git a/packages/ludic.render3d/r3d.ludic b/packages/ludic.render3d/r3d.ludic
index 135ce808..55fa9d89 100644
--- a/packages/ludic.render3d/r3d.ludic
+++ b/packages/ludic.render3d/r3d.ludic
@@ -31,6 +31,7 @@ import "scatter.ludic"
 import "actor.ludic"
 import "stream.ludic"
 import "grass.ludic"
+import "grass_gpu.ludic"
 import "water.ludic"
 import "streamline.ludic"
 import "render.ludic"
diff --git a/packages/ludic.render3d/render.ludic b/packages/ludic.render3d/render.ludic
index 0fcd601e..fb98638f 100644
--- a/packages/ludic.render3d/render.ludic
+++ b/packages/ludic.render3d/render.ludic
@@ -58,6 +58,7 @@ function r3d_env_flags(render3d_st: mut Render3dState) -> void {
   render3d_st.sc_skip_blade = r3d_env_has(render3d_st, "R3D_NOBLADES")
   render3d_st.sc_skip_grass = r3d_env_has(render3d_st, "R3D_NOGRASS")
   render3d_st.grass_force = r3d_env_has(render3d_st, "R3D_BLADES")
+  render3d_st.grass_env_off = r3d_env_has(render3d_st, "R3D_NOBLADES")
   render3d_st.sc_skip_card = r3d_env_has(render3d_st, "R3D_NOCARDS")
   render3d_st.sc_dbg_lod = r3d_env_has(render3d_st, "R3D_LODDBG")
   if r3d_env_has(render3d_st, "R3D_ANISO") {
@@ -208,6 +209,7 @@ function r3d_frame(render3d_st: mut Render3dState, time: float) -> void {
   prof_cpu_mark(render3d_st, "shadow rebake")
   stream_update_all(render3d_st)
   prof_cpu_mark(render3d_st, "streaming")
+  gg_cull_frame(render3d_st)                # before any pass: a dispatch inside one would split it
   if not render3d_st.r3d_no_shadow { prof_begin(render3d_st, "shadow");  shadow_pass(render3d_st);  prof_end(render3d_st) }
   prof_cpu_mark(render3d_st, "shadow pass")
   if render3d_st.water_on and not render3d_st.r3d_no_refl and water_reflect_visible(render3d_st) { prof_begin(render3d_st, "water reflection");  water_reflection_pass(render3d_st);  prof_end(render3d_st) }
diff --git a/packages/ludic.render3d/scatter.ludic b/packages/ludic.render3d/scatter.ludic
index 005549b3..2f24fe8d 100644
--- a/packages/ludic.render3d/scatter.ludic
+++ b/packages/ludic.render3d/scatter.ludic
@@ -36,7 +36,7 @@ property Layer {
   last_cam: floats,
   rough:    float = 0.0,
   blade:    bool = false,
-  grass:    bool = false,        # ground-cover grass the game may hand to the GPU blades (R3D_NOGRASS skips it)
+  grass:    bool = false,        # ground cover the GPU blades replace: skipped while they draw (and under R3D_NOGRASS)
   flower:   bool = false,
   card:     bool = false,
   cheap:    bool = false,      # distant cover: no shadows, no wind, flat lighting
@@ -1135,7 +1135,7 @@ function scatter_draw_depth(render3d_st: mut Render3dState) -> void {
   for i in 0 .. len(render3d_st.sc_layers) {
     let l = render3d_st.sc_layers[i]
     if not l.foliage or l.blade or l.flower { continue }
-    if render3d_st.sc_skip_grass and l.grass { continue }
+    if l.grass and sc_grass_replaced(render3d_st) { continue }
     if render3d_st.r3d_no_trees and l.imp != null { continue }
     layer_update(render3d_st, l)
     if l.n_lods > 1 and l.g_on {
@@ -1151,13 +1151,15 @@ function scatter_draw_depth(render3d_st: mut Render3dState) -> void {
   }
   gpu_cull(render3d_st, true)
 }
+# a layer the game flagged `grass` is ground cover the GPU blades stand in for: off while they draw
+function sc_grass_replaced(render3d_st: Render3dState) -> bool { return render3d_st.sc_skip_grass or ((render3d_st.grass_on or render3d_st.grass_force) and not render3d_st.grass_env_off and render3d_st.grass_prog != 0) }
 function scatter_draw(render3d_st: mut Render3dState) -> void {
   for i in 0 .. len(render3d_st.sc_layers) {
     let l = render3d_st.sc_layers[i]
     if render3d_st.sc_skip_blade and l.blade { continue }
     if render3d_st.sc_skip_flower and l.flower { continue }
     if render3d_st.sc_skip_card and l.card { continue }
-    if render3d_st.sc_skip_grass and l.grass { continue }
+    if l.grass and sc_grass_replaced(render3d_st) { continue }
     if render3d_st.r3d_no_trees and l.imp != null and not l.card { continue }
     layer_update(render3d_st, l)
     layer_draw_near(render3d_st, l, false, null, false)
@@ -1175,7 +1177,7 @@ function scatter_draw_casters(render3d_st: mut Render3dState, light_vp: floats)
     let l = render3d_st.sc_layers[i]
     if render3d_st.sc_skip_blade and l.blade { continue }
     if render3d_st.sc_skip_card and l.card { continue }
-    if render3d_st.sc_skip_grass and l.grass { continue }
+    if l.grass and sc_grass_replaced(render3d_st) { continue }
     if render3d_st.r3d_no_trees and l.imp != null and not l.card { continue }
     layer_update(render3d_st, l)
     if l.imp != null {
diff --git a/packages/ludic.render3d/shaders/compute.list b/packages/ludic.render3d/shaders/compute.list
index 10bd7bd4..fafa1bcb 100644
--- a/packages/ludic.render3d/shaders/compute.list
+++ b/packages/ludic.render3d/shaders/compute.list
@@ -2,3 +2,5 @@
 # spv/<name>.comp.spv. Binding 0 is the uniform block of parameters, 1.. storage buffers.
 probe|probe.comp
 scatter_cull|scatter_cull.comp
+grass_reset|grass_reset.comp
+grass_cull|grass_cull.comp
diff --git a/packages/ludic.render3d/shaders/grass.mesh b/packages/ludic.render3d/shaders/grass.mesh
index 60f50ce0..b172254d 100644
--- a/packages/ludic.render3d/shaders/grass.mesh
+++ b/packages/ludic.render3d/shaders/grass.mesh
@@ -31,6 +31,7 @@ uniform vec4 u_tiles[256];   // per tile of the dispatch: corner x, corner z, in
 uniform float u_s0;
 uniform float u_d0;
 uniform float u_radius;
+uniform float u_px;          // one pixel in radians (grass.vert)
 uniform int u_dbg;
 out vec3 v_wpos[];
 out vec3 v_nrm[];
@@ -40,7 +41,7 @@ out vec2 v_rot[];
 out float v_hull[];
 out float v_quake[];   // grass does not quake; written so model.frag can read it
 
-const float CELL = 16.0;
+const float CELL = 4.0;   // grass.ludic GRASS_CELL: 4 m, so a near tile asks for what its blades need
 float hash1(vec2 p) { return fract(sin(dot(p, vec2(127.1, 311.7))) * 43758.5453123); }
 uint pcg(uint v) { uint s = v * 747796405u + 2891336453u; uint w = ((s >> ((s >> 28u) + 4u)) ^ s) * 277803737u; return (w >> 22u) ^ w; }
 float bladeHash(ivec2 cell, int j, int k) {
@@ -115,6 +116,7 @@ void main() {
     float tall = mix(0.18, 0.42, h3) * mix(0.8, 1.2, hash1(cid * 0.1)) * (1.0 + 0.35 * smoothstep(1.0, 12.0, grow)) * life;
     float bw = 0.028 * mix(1.0, 0.45 * grow, smoothstep(1.0, 4.0, grow));
     if (far) { bw = max(bw, spacing * 0.35); tall = min(tall, spacing * 0.3); }
+    bw = max(bw, dist * u_px * 1.1);
     float gust = sin(xz.x * 0.09 + u_time * 1.1) * 0.5 + sin(xz.y * 0.13 - u_time * 0.8 + xz.x * 0.05) * 0.5;
     float ph = u_time * 1.7 + seed * 6.2831 + xz.x * 0.05 + xz.y * 0.07;
     float sway = (sin(ph) * 0.6 + sin(ph * 2.3 + 1.0) * 0.4 + gust) * u_wind;
diff --git a/packages/ludic.render3d/shaders/grass.vert b/packages/ludic.render3d/shaders/grass.vert
index 03475942..32f75c16 100644
--- a/packages/ludic.render3d/shaders/grass.vert
+++ b/packages/ludic.render3d/shaders/grass.vert
@@ -34,6 +34,7 @@ uniform vec4 u_tiles[256];   // per tile of the draw: corner x, corner z, indice
 uniform float u_s0;          // blade spacing at the camera (m)
 uniform float u_d0;          // distance at which the spacing has doubled (m)
 uniform float u_radius;      // no blades past this
+uniform float u_px;          // one pixel's height in radians: no blade is drawn narrower than a pixel
 uniform int u_dbg;           // R3D_GRASS_DBG: 1 lift blades 0.3 m, 2 light as ground everywhere, 3 both
 out vec3 v_wpos;
 out vec3 v_nrm;
@@ -42,9 +43,10 @@ out float v_seed;
 out vec2 v_rot;
 out float v_hull;
 out float v_quake;   // only aspens quake; written so model.frag can read it
+out vec3 v_tint;     // the blade's colour field at its root (model.frag's regionTint and patchiness, once a blade)
 
-const float CELL = 16.0;
-float hash1(vec2 p) { return fract(sin(dot(p, vec2(127.1, 311.7))) * 43758.5453123); }
+const float CELL = 4.0;   // grass.ludic GRASS_CELL: 4 m, so a near tile asks for what its blades need
+// hash1 and fbm come from noise.glsl, which a GBLADE vertex stage is given (programs.ludic)
 // An integer hash (PCG) for the per-blade values. The sine hash advanced linearly with
 // the blade index, so a cell's blades fell into diagonal rows, and the rows read as
 // streaks across the meadow with an edge wherever they thinned out.
@@ -68,7 +70,22 @@ float heightSmooth(sampler2D tex, vec2 uv) {
   return (texture(tex, vec2(o0.x, o0.y)).r * s0.x + texture(tex, vec2(o1.x, o0.y)).r * s1.x) * s0.y
        + (texture(tex, vec2(o0.x, o1.y)).r * s0.x + texture(tex, vec2(o1.x, o1.y)).r * s1.x) * s1.y;
 }
-void cull() { gl_Position = vec4(0.0, 0.0, 2.0, 1.0); v_wpos = vec3(0.0); v_nrm = vec3(0.0, 1.0, 0.0); v_uv = vec2(0.0); v_seed = 0.0; v_rot = vec2(0.0, 1.0); v_hull = 1.0; v_quake = 0.0; }
+void cull() { gl_Position = vec4(0.0, 0.0, 2.0, 1.0); v_wpos = vec3(0.0); v_nrm = vec3(0.0, 1.0, 0.0); v_uv = vec2(0.0); v_seed = 0.0; v_rot = vec2(0.0, 1.0); v_hull = 1.0; v_quake = 0.0; v_tint = vec3(1.0); }
+// lighting.glsl's regionTint(wpos, 0.35) times model.frag's patchiness, at the root: three fbm
+// fields a blade pixel used to pay for, when the whole blade stands on one spot
+vec3 bladeField(vec2 xz, float y) {
+  float n = fbm(xz * 0.0018 + 4.0, 3) * 0.5 + 0.5;
+  float aspen = smoothstep(0.52, 0.7, n) * smoothstep(520.0, 250.0, y);
+  float dry = smoothstep(0.35, 0.15, fbm(xz * 0.004 + 9.0, 3) * 0.5 + 0.5);
+  vec3 t = mix(vec3(1.0), vec3(1.25, 1.3, 0.85), aspen * 0.35);
+  t = mix(t, vec3(1.15, 1.05, 0.7), dry * 0.35 * 0.6);
+  float patchy = fbm(xz * 0.045, 3) * 0.5 + 0.5;
+  t *= mix(vec3(0.7, 0.8, 0.55), vec3(1.1, 1.05, 0.85), patchy);
+  // dry and green patches a few metres across, and the tussocks' own shade of the same green
+  float dryp = smoothstep(0.2, 0.8, gnoise(xz * 0.33 + 17.0) * 0.5 + 0.5);
+  t *= mix(vec3(0.92, 1.0, 0.95), vec3(1.30, 1.10, 0.62), dryp);
+  return t * (0.72 + 0.56 * (gnoise(xz * 1.9 + 41.0) * 0.5 + 0.5));
+}
 
 void main() {
 #ifdef TILES
@@ -158,6 +175,10 @@ void main() {
   // that had been cut, which is the one thing an alpine meadow is not.
   float hh = h3 * h3;
   float tall = mix(0.09, 0.60, hh) * mix(0.8, 1.2, hash1(cid * 0.1)) * (1.0 + 0.35 * smoothstep(1.0, 12.0, grow)) * life;
+  // CLUMPS: grass grows in tussocks, taller and shorter patches a metre or so across, and that
+  // relief is the texture a meadow has from above - an even height reads as felt
+  float clump = gnoise(xz * 0.85 + 3.7) * 0.5 + 0.5;
+  tall *= mix(0.45, 1.40, clump * clump * (3.0 - 2.0 * clump));
   // 2.8 cm was FIVE TIMES a blade of meadow grass, which is 3-6 mm. At 2 m from the
   // camera that is a broad dark scimitar lying along the ground rather than a blade
   // standing in a sward, and no amount of profile or colour work fixes a blade that is
@@ -165,6 +186,9 @@ void main() {
   // and too flat" note in this stage.
   float bw = 0.010 * mix(1.0, 0.45 * grow, smoothstep(1.0, 4.0, grow));
   if (far) { bw = max(bw, spacing * 0.35); tall = min(tall, spacing * 0.3); }
+  // a blade under a pixel is a vertex bill that shimmers: widen it to one and a bit, and let the
+  // thinning (spacing) keep the coverage
+  bw = max(bw, dist * u_px * 1.1);
   vec3 p = vec3(a_pos.x * bw, a_pos.y * tall, a_pos.z * tall * (0.6 + 0.8 * h4));
   vec3 n = vec3(0.0, 0.3, 1.0);
   // The shared field (wind.glsl), so the gust that crosses this meadow is the same gust that
@@ -217,6 +241,7 @@ void main() {
   v_seed = seed;
   v_rot = vec2(s, c_);
   v_quake = 0.0;
+  v_tint = bladeField(xz, h);
   v_hull = (dist > 2.0 || far || (u_dbg & 2) != 0) ? -1.0 : 1.0;   // no back-face flip, no rounding past arm's reach
   gl_Position = u_proj * u_view * vec4(w, 1.0);
 }
diff --git a/packages/ludic.render3d/shaders/grass_cull.comp b/packages/ludic.render3d/shaders/grass_cull.comp
new file mode 100644
index 00000000..4f4567ad
--- /dev/null
+++ b/packages/ludic.render3d/shaders/grass_cull.comp
@@ -0,0 +1,123 @@
+// grass_cull.comp - the meadow's blades, decided once a frame on the GPU (Vulkan).
+//
+// grass.vert used to do all of this per VERTEX: every blade's hashes, its place, the ground under
+// it (a four-tap bicubic height), the photograph's say on whether anything grows there, the water,
+// the slope, the snow line and its colour field - ten times over for a five-row blade, and in full
+// for every index a tile asked for and then threw away. Here each candidate is one invocation: a
+// workgroup row per visible tile (grass.ludic walks them as before), a thread per candidate blade.
+// A blade that survives is written once, as four vec4s, into its band's region, and the band's
+// indirect draw count goes up by one; grass_inst.vert only bends and places the vertices.
+// The rules are grass.vert's, kept in step with it: the same hashes give the same meadow.
+layout(local_size_x = 64) in;
+
+layout(set = 0, binding = 0) uniform Params {
+  vec4 planes[4];   // the side frustum planes: xyz in, w distance
+  vec4 cam;         // xyz the camera, w no blades past this (u_radius)
+  vec4 dens;        // x s0, y d0, z one pixel in radians, w the snow line
+  vec4 lake;        // the carved lake: centre x/z, half extents (z = 0: none)
+  vec4 lev;         // x the lake's level, y the sea's, z the photograph on (1), w unused
+  vec4 ts;          // the height texture: origin x/z, half extent, unused
+  vec4 bands;       // outer distance of bands 0 and 1 (x, y); band 2 is the rest
+  uvec4 base;       // each band's first record in Out
+  uvec4 cap;        // ... and how many it holds
+} pr;
+
+layout(set = 0, binding = 1) readonly buffer Tiles { vec4 tiles[]; };   // corner x/z, indices per cell, cells per side
+layout(set = 0, binding = 2) buffer Out { vec4 outv[]; };               // 4 per blade
+layout(set = 0, binding = 3) buffer Cmds { uint cmds[]; };              // VkDrawIndexedIndirectCommand per band
+layout(set = 0, binding = 4) uniform sampler2D u_height;                // height, normal (terrain's u_ts_height)
+layout(set = 0, binding = 5) uniform sampler2D u_ortho;                 // the photograph
+
+const float CELL = 4.0;   // grass.ludic GRASS_CELL
+
+uint pcg(uint v) { uint s = v * 747796405u + 2891336453u; uint w = ((s >> ((s >> 28u) + 4u)) ^ s) * 277803737u; return (w >> 22u) ^ w; }
+float bladeHash(ivec2 cell, int j, int k) {
+  uint h = pcg(uint(cell.x + 32768) * 73856093u ^ uint(cell.y + 32768) * 19349663u ^ uint(j) * 83492791u ^ uint(k) * 2654435761u);
+  return float(h) * (1.0 / 4294967295.0);
+}
+float heightSmooth(vec2 uv) {
+  vec2 res = vec2(textureSize(u_height, 0));
+  vec2 t = uv * res - 0.5;
+  vec2 f = fract(t);
+  vec2 i = floor(t);
+  vec2 w0 = (1.0 - f) * (1.0 - f) * (1.0 - f) / 6.0;
+  vec2 w1 = (4.0 - 6.0 * f * f + 3.0 * f * f * f) / 6.0;
+  vec2 w3 = f * f * f / 6.0;
+  vec2 w2 = 1.0 - w0 - w1 - w3;
+  vec2 s0 = w0 + w1, s1 = w2 + w3;
+  vec2 o0 = (i - 1.0 + w1 / s0 + 0.5) / res;
+  vec2 o1 = (i + 1.0 + w3 / s1 + 0.5) / res;
+  return (textureLod(u_height, vec2(o0.x, o0.y), 0.0).r * s0.x + textureLod(u_height, vec2(o1.x, o0.y), 0.0).r * s1.x) * s0.y
+       + (textureLod(u_height, vec2(o0.x, o1.y), 0.0).r * s0.x + textureLod(u_height, vec2(o1.x, o1.y), 0.0).r * s1.x) * s1.y;
+}
+// grass.vert's bladeField: the region, the patchiness, the dry patches and the tussocks' shade
+vec3 bladeField(vec2 xz, float y) {
+  float n = fbm(xz * 0.0018 + 4.0, 3) * 0.5 + 0.5;
+  float aspen = smoothstep(0.52, 0.7, n) * smoothstep(520.0, 250.0, y);
+  float dry = smoothstep(0.35, 0.15, fbm(xz * 0.004 + 9.0, 3) * 0.5 + 0.5);
+  vec3 t = mix(vec3(1.0), vec3(1.25, 1.3, 0.85), aspen * 0.35);
+  t = mix(t, vec3(1.15, 1.05, 0.7), dry * 0.35 * 0.6);
+  float patchy = fbm(xz * 0.045, 3) * 0.5 + 0.5;
+  t *= mix(vec3(0.7, 0.8, 0.55), vec3(1.1, 1.05, 0.85), patchy);
+  float dryp = smoothstep(0.2, 0.8, gnoise(xz * 0.33 + 17.0) * 0.5 + 0.5);
+  t *= mix(vec3(0.92, 1.0, 0.95), vec3(1.30, 1.10, 0.62), dryp);
+  return t * (0.72 + 0.56 * (gnoise(xz * 1.9 + 41.0) * 0.5 + 0.5));
+}
+
+void main() {
+  vec4 tile = tiles[gl_WorkGroupID.y];
+  int per_cell = int(tile.z + 0.5);
+  int cells = int(tile.w + 0.5);
+  int i = int(gl_GlobalInvocationID.x);
+  if (per_cell <= 0 || i >= per_cell * cells * cells) return;
+  int c = i / per_cell;
+  int j = i - c * per_cell;
+  vec2 cell = tile.xy + vec2(float(c % cells), float(c / cells)) * CELL;
+  vec2 cid = floor(cell / CELL + 0.5);
+  ivec2 ci = ivec2(cid);
+  vec2 hv = vec2(bladeHash(ci, j, 0), bladeHash(ci, j, 1));
+  vec2 xz = cell + hv * CELL;
+  float dist = length(xz - pr.cam.xz);
+  float radius = pr.cam.w;
+  if (dist >= radius) return;
+  float spacing = pr.dens.x * (1.0 + dist / pr.dens.y);
+  float count = CELL * CELL / (spacing * spacing) * (1.0 - smoothstep(radius * 0.7, radius, dist));
+  float keep = count / float(max(per_cell, 1));
+  float r = bladeHash(ci, j, 5);
+  if (r > keep) return;
+  float life = 1.0 - smoothstep(keep * 0.75, keep, r);
+  vec2 huv = (xz - pr.ts.xy) / (2.0 * pr.ts.z) + 0.5;
+  if (huv.x < 0.0 || huv.x > 1.0 || huv.y < 0.0 || huv.y > 1.0) return;
+  vec4 ht = textureLod(u_height, huv, 0.0);
+  // the frustum, on a sphere round the blade (it stands at most 0.9 m tall)
+  vec3 root = vec3(xz.x, ht.r, xz.y);
+  for (int k = 0; k < 4; k++) { if (dot(pr.planes[k].xyz, root) + pr.planes[k].w < -1.0) return; }
+  vec3 gn = normalize(ht.gba);
+  float h3 = bladeHash(ci, j, 2), h4 = bladeHash(ci, j, 3);
+  float wl = pr.lev.y;
+  if (pr.lake.z > 0.0) { vec2 q = (xz - pr.lake.xy) / pr.lake.zw;  if (dot(q, q) < 1.0) wl = max(wl, pr.lev.x); }
+  float ok = (1.0 - smoothstep(0.30, 0.55, 1.0 - gn.y)) * smoothstep(0.0, 0.6, ht.r - wl - 0.15) * smoothstep(pr.dens.w - 80.0, pr.dens.w - 200.0, ht.r);
+  if (pr.lev.z > 0.5) {
+    vec3 oc = textureLod(u_ortho, huv, 1.5).rgb;
+    ok *= 0.40 + 0.60 * smoothstep(0.0, 0.025, oc.g - oc.b);
+  }
+  if (h4 > ok) return;
+  float h = heightSmooth(huv);
+  float seed = hv.x * 0.7 + hv.y * 0.3;
+  float ang = bladeHash(ci, j, 6) * 6.2831853;
+  float grow = spacing / pr.dens.x;
+  float hh = h3 * h3;
+  float tall = mix(0.09, 0.60, hh) * mix(0.8, 1.2, hash1(cid * 0.1)) * (1.0 + 0.35 * smoothstep(1.0, 12.0, grow)) * life;
+  float clump = gnoise(xz * 0.85 + 3.7) * 0.5 + 0.5;
+  tall *= mix(0.45, 1.40, clump * clump * (3.0 - 2.0 * clump));
+  float bw = 0.010 * mix(1.0, 0.45 * grow, smoothstep(1.0, 4.0, grow));
+  bw = max(bw, dist * pr.dens.z * 1.1);
+  uint b = dist < pr.bands.x ? 0u : (dist < pr.bands.y ? 1u : 2u);
+  uint slot = atomicAdd(cmds[b * 5u + 1u], 1u);
+  if (slot >= pr.cap[b]) { atomicAdd(cmds[b * 5u + 1u], 0xFFFFFFFFu); return; }
+  uint o = (pr.base[b] + slot) * 4u;
+  outv[o] = vec4(xz.x, h - 0.02, xz.y, seed);
+  outv[o + 1u] = vec4(sin(ang), cos(ang), tall, bw);
+  outv[o + 2u] = vec4(bladeField(xz, h), 0.6 + 0.8 * h4);
+  outv[o + 3u] = vec4(gn, dist);
+}
diff --git a/packages/ludic.render3d/shaders/grass_inst.vert b/packages/ludic.render3d/shaders/grass_inst.vert
new file mode 100644
index 00000000..c501ed90
--- /dev/null
+++ b/packages/ludic.render3d/shaders/grass_inst.vert
@@ -0,0 +1,71 @@
+// grass_inst.vert - a blade grass_cull.comp has already decided on (Vulkan). Everything about the
+// blade - where it stands, its facing, height, width, colour field and the ground under it - is its
+// instance record; this only bends, sways, pushes and stands the vertices, as grass.vert's second
+// half does, so the two draw the same meadow.
+layout(location = 0) in vec3 a_pos;      // x: -0.5..0.5 across, y: 0..1 along the blade, z: bend
+layout(location = 2) in vec2 a_uv;
+layout(location = 3) in vec4 i_root;     // xyz the root, w the seed
+layout(location = 4) in vec4 i_shape;    // sin and cos of the yaw, height, width
+layout(location = 5) in vec4 i_tint;     // the colour field, and how far the tip arches
+layout(location = 6) in vec4 i_ground;   // the ground's normal, and the distance it was decided at
+uniform mat4 u_view;
+uniform mat4 u_proj;
+uniform float u_time;
+uniform float u_wind;
+uniform vec3 u_push;      // x, z, radius: a body standing in the grass
+uniform int u_dbg;        // R3D_GRASS_DBG (2: light as ground everywhere)
+out vec3 v_wpos;
+out vec3 v_nrm;
+out vec2 v_uv;
+out float v_seed;
+out vec2 v_rot;
+out float v_hull;
+out float v_quake;
+out vec3 v_tint;
+
+void main() {
+  vec2 xz = i_root.xz;
+  float seed = i_root.w;
+  float s = i_shape.x, c_ = i_shape.y, tall = i_shape.z, bw = i_shape.w;
+  vec3 gn = i_ground.xyz;
+  float dist = i_ground.w;
+  vec3 p = vec3(a_pos.x * bw, a_pos.y * tall, a_pos.z * tall * i_tint.w);
+  vec3 n = vec3(0.0, 0.3, 1.0);
+  float ph = u_time * 1.7 + seed * 6.2831 + xz.x * 0.05 + xz.y * 0.07;
+  float sway = windSway(xz, u_time, ph, 0.0) * u_wind;
+  float hgt = max(p.y, 0.0);
+  p.x += sway * hgt * hgt * 0.35;
+  p.z += sway * hgt * hgt * 0.15 * cos(ph * 0.7);
+  p = vec3(c_ * p.x + s * p.z, p.y, -s * p.x + c_ * p.z);
+  if (u_push.z > 0.0) {
+    vec2 away = xz - u_push.xy;
+    float pd = length(away);
+    float push = smoothstep(u_push.z, u_push.z * 0.2, pd);
+    if (push > 0.0) {
+      vec2 pdir = (pd > 1e-3) ? away / pd : vec2(1.0, 0.0);
+      float bh = max(p.y, 0.0);
+      p.x += pdir.x * push * bh * 1.05;
+      p.z += pdir.y * push * bh * 1.05;
+      p.y -= push * bh * 0.40;
+    }
+  }
+  n = normalize(vec3(c_ * n.x + s * n.z, n.y, -s * n.x + c_ * n.z));
+  vec3 k = cross(vec3(0.0, 1.0, 0.0), gn);
+  float sk = length(k), ck = gn.y;
+  if (sk > 1e-4) {
+    k /= sk;
+    p = p * ck + cross(k, p) * sk + k * dot(k, p) * (1.0 - ck);
+    n = normalize(n * ck + cross(k, n) * sk + k * dot(k, n) * (1.0 - ck));
+  }
+  n = normalize(mix(n, gn, smoothstep(2.0, 12.0, dist)));
+  vec3 w = i_root.xyz + p;
+  v_wpos = w;
+  v_nrm = n;
+  v_uv = a_uv;
+  v_seed = seed;
+  v_rot = vec2(s, c_);
+  v_quake = 0.0;
+  v_tint = i_tint.rgb;
+  v_hull = (dist > 2.0 || (u_dbg & 2) != 0) ? -1.0 : 1.0;
+  gl_Position = u_proj * u_view * vec4(w, 1.0);
+}
diff --git a/packages/ludic.render3d/shaders/grass_reset.comp b/packages/ludic.render3d/shaders/grass_reset.comp
new file mode 100644
index 00000000..f810b0bc
--- /dev/null
+++ b/packages/ludic.render3d/shaders/grass_reset.comp
@@ -0,0 +1,8 @@
+// grass_reset.comp - zero each band's instance count before grass_cull.comp counts into it. On the
+// GPU, in order, because the frame before may still be drawing from the same records.
+layout(local_size_x = 1) in;
+layout(set = 0, binding = 0) uniform Params { uvec4 n; } pr;
+layout(set = 0, binding = 1) buffer Cmds { uint cmds[]; };
+void main() {
+  for (uint b = 0u; b < pr.n.x; b++) { cmds[b * 5u + 1u] = 0u; }
+}
diff --git a/packages/ludic.render3d/shaders/model.frag b/packages/ludic.render3d/shaders/model.frag
index 41257a24..d6356ab3 100644
--- a/packages/ludic.render3d/shaders/model.frag
+++ b/packages/ludic.render3d/shaders/model.frag
@@ -15,6 +15,9 @@ uniform mat4 u_view;
 uniform vec3 u_tint;
 uniform float u_rough_scale;
 uniform float u_emissive;     // self-lit (a flame): albedo added back after shading
+#ifdef GBLADE
+in vec3 v_tint;               // grass.vert's colour field (the mesh-shader and scatter blades work it out here)
+#endif
 #ifdef BLADE
 uniform vec3 u_blade_base;
 uniform vec3 u_blade_tip;
@@ -49,6 +52,11 @@ void main() {
 #endif
   vec3 N = normalize(v_nrm);
   if (!gl_FrontFacing && v_hull >= 0.0) N = -N;
+#ifdef BLADE
+  // A blade is thin and two-sided: the sun lights whichever face it reaches (the other glows
+  // through, below). Faced away from the sun, half the meadow was lit by the sky alone.
+  if (dot(N, u_sun_dir) < 0.0) N = -N;
+#endif
 #ifdef CARD
   // a baked card: albedo with coverage (premultiplied mips), normal in the card's own frame
   vec4 ca = texture(u_diff, v_uv);
@@ -80,7 +88,16 @@ void main() {
 #ifdef BLADE
   // a procedural blade: dark at the root, lighter at the tip; some blades gone to seed
   float t = clamp(v_uv.y, 0.0, 1.0);
-  alb = mix(u_blade_base, u_blade_tip, t * t) * regionTint(v_wpos, 0.35);
+  // most of a blade is its own green, only the sheath darker: t*t held the lower two thirds at the
+  // root colour (0.14, darker than the ground), and a meadow of those reads as dark wire
+#ifndef GBLADE
+  vec3 field = regionTint(v_wpos, 0.35);
+  float patchy = fbm(v_wpos.xz * 0.045, 3) * 0.5 + 0.5;
+  field *= mix(vec3(0.7, 0.8, 0.55), vec3(1.1, 1.05, 0.85), patchy);
+#else
+  vec3 field = v_tint;
+#endif
+  alb = mix(u_blade_base, u_blade_tip, smoothstep(0.0, 0.55, t)) * field;
   float dry = smoothstep(0.7, 0.8, fract(v_seed * 3.17));
   alb = mix(alb, vec3(0.40, 0.36, 0.13) * (0.45 + 0.55 * t), dry * 0.75);
   // A REAL BLADE, if the game gave us one. Each blade picks a column of the atlas from
@@ -110,13 +127,13 @@ void main() {
     vec3 g = photo * (1.0 / 0.3736);
     alb *= mix(vec3(1.0), min(g, vec3(1.15)), 0.85);
   }
-  float patchy = fbm(v_wpos.xz * 0.045, 3) * 0.5 + 0.5;
-  alb *= mix(vec3(0.7, 0.8, 0.55), vec3(1.1, 1.05, 0.85), patchy);
   // a far tuft is a patch of the meadow, darker than a lit blade tip and never straw
-  if (v_hull < 0.0) alb = mix(u_blade_base, u_blade_tip, 0.45) * regionTint(v_wpos, 0.35) * mix(vec3(0.7, 0.8, 0.55), vec3(1.1, 1.05, 0.85), patchy) * 0.72;
+  if (v_hull < 0.0) alb = mix(u_blade_base, u_blade_tip, 0.45) * field * 0.72;
   // a rounded cross-section reads softer than a flat card
   vec3 side = normalize(cross(N, vec3(0.0, 1.0, 0.0)) + vec3(1e-4));
   n = (v_hull < 0.0) ? N : normalize(N + side * (v_uv.x * 2.0 - 1.0) * 0.6);
+  // and a sward is lit as a mass leaning to the sky: a level normal took the light at a graze
+  n = normalize(n + vec3(0.0, 0.45, 0.0));
   // A blade's own ambient occlusion. This was mix(0.2, 1.0, t*t): 80% occluded at the
   // sheath and still 60% at half height, because t*t holds the curve down. Measured in a
   // walking-distance shot the blades came out at 24-31 of 255 against a ground of 143 -
@@ -227,6 +244,13 @@ void main() {
 #else
   float shadow = mix(1.0, sunShadow(v_wpos + N * 0.1 + vec3(0.0, 0.2, 0.0), N, viewDepth), 0.55);
 #endif
+#elif defined(BLADE)
+  // A blade stands within a shadow texel of the ground under it, and biased along its own
+  // (level) normal it read the terrain as its caster: the whole meadow in shadow, dark wire.
+  // Looked up a hand's width up, along the ground's up, trees and rocks still shade it.
+  // past arm's length a blade is a pixel or two wide: one tap is what a soft filter averages to
+  float shadow = (dist < 8.0) ? sunShadow(v_wpos + vec3(0.0, 0.25, 0.0), vec3(0.0, 1.0, 0.0), viewDepth)
+                              : sunShadowCheap(v_wpos + vec3(0.0, 0.25, 0.0), vec3(0.0, 1.0, 0.0), viewDepth);
 #else
   float shadow = sunShadow(v_wpos, N, viewDepth);
 #endif
@@ -271,6 +295,9 @@ void main() {
     float NoV = max(dot(n, vv), 0.0);
     float sheen = pow(NoH, 26.0) * 0.55 + pow(1.0 - NoV, 4.0) * 0.05;
     sheen *= smoothstep(0.10, 0.42, NoV);
+#ifdef BLADE
+    sheen *= 0.2;   // a sward seen from above is tips edge-on to the sun: the leaf's glint frosted them white
+#endif
     col += u_sun_color * sheen * shadow * cloudShadow(v_wpos) * 0.4 * (0.45 + 0.55 * arm.r);
   }
 #endif
diff --git a/packages/ludic.render3d/shaders/spv/7ed7de52.frag.spv b/packages/ludic.render3d/shaders/spv/7ed7de52.frag.spv
deleted file mode 100644
index b54d682d..00000000
Binary files a/packages/ludic.render3d/shaders/spv/7ed7de52.frag.spv and /dev/null differ
diff --git a/packages/ludic.render3d/shaders/spv/7ed7de52.vert.spv b/packages/ludic.render3d/shaders/spv/7ed7de52.vert.spv
deleted file mode 100644
index 5fd3b45c..00000000
Binary files a/packages/ludic.render3d/shaders/spv/7ed7de52.vert.spv and /dev/null differ
diff --git a/packages/ludic.render3d/shaders/spv/7f096322.frag.spv b/packages/ludic.render3d/shaders/spv/7f096322.frag.spv
deleted file mode 100644
index b54d682d..00000000
Binary files a/packages/ludic.render3d/shaders/spv/7f096322.frag.spv and /dev/null differ
diff --git a/packages/ludic.render3d/shaders/spv/7f096322.vert.spv b/packages/ludic.render3d/shaders/spv/7f096322.vert.spv
deleted file mode 100644
index eb02b0d0..00000000
Binary files a/packages/ludic.render3d/shaders/spv/7f096322.vert.spv and /dev/null differ
diff --git a/packages/ludic.render3d/shaders/spv/b20fdcb5.frag.spv b/packages/ludic.render3d/shaders/spv/b20fdcb5.frag.spv
new file mode 100644
index 00000000..35eaa795
Binary files /dev/null and b/packages/ludic.render3d/shaders/spv/b20fdcb5.frag.spv differ
diff --git a/packages/ludic.render3d/shaders/spv/b20fdcb5.vert.spv b/packages/ludic.render3d/shaders/spv/b20fdcb5.vert.spv
new file mode 100644
index 00000000..3744bcf9
Binary files /dev/null and b/packages/ludic.render3d/shaders/spv/b20fdcb5.vert.spv differ
diff --git a/packages/ludic.render3d/shaders/spv/b8cff226.frag.spv b/packages/ludic.render3d/shaders/spv/b8cff226.frag.spv
new file mode 100644
index 00000000..a7862f95
Binary files /dev/null and b/packages/ludic.render3d/shaders/spv/b8cff226.frag.spv differ
diff --git a/packages/ludic.render3d/shaders/spv/b8cff226.vert.spv b/packages/ludic.render3d/shaders/spv/b8cff226.vert.spv
new file mode 100644
index 00000000..12eafb74
Binary files /dev/null and b/packages/ludic.render3d/shaders/spv/b8cff226.vert.spv differ
diff --git a/packages/ludic.render3d/shaders/spv/cb13f3c6.frag.spv b/packages/ludic.render3d/shaders/spv/cb13f3c6.frag.spv
new file mode 100644
index 00000000..a7862f95
Binary files /dev/null and b/packages/ludic.render3d/shaders/spv/cb13f3c6.frag.spv differ
diff --git a/packages/ludic.render3d/shaders/spv/cb13f3c6.vert.spv b/packages/ludic.render3d/shaders/spv/cb13f3c6.vert.spv
new file mode 100644
index 00000000..4858c099
Binary files /dev/null and b/packages/ludic.render3d/shaders/spv/cb13f3c6.vert.spv differ
diff --git a/packages/ludic.render3d/shaders/spv/d48d6a48.frag.spv b/packages/ludic.render3d/shaders/spv/d48d6a48.frag.spv
index b54d682d..39de29f8 100644
Binary files a/packages/ludic.render3d/shaders/spv/d48d6a48.frag.spv and b/packages/ludic.render3d/shaders/spv/d48d6a48.frag.spv differ
diff --git a/packages/ludic.render3d/shaders/spv/d48d6a48.vert.spv b/packages/ludic.render3d/shaders/spv/d48d6a48.vert.spv
index 7b7b8aa1..0749c351 100644
Binary files a/packages/ludic.render3d/shaders/spv/d48d6a48.vert.spv and b/packages/ludic.render3d/shaders/spv/d48d6a48.vert.spv differ
diff --git a/packages/ludic.render3d/shaders/spv/f4a2b586.frag.spv b/packages/ludic.render3d/shaders/spv/f4a2b586.frag.spv
index 071188aa..21e807a8 100644
Binary files a/packages/ludic.render3d/shaders/spv/f4a2b586.frag.spv and b/packages/ludic.render3d/shaders/spv/f4a2b586.frag.spv differ
diff --git a/packages/ludic.render3d/shaders/spv/grass_cull.comp.spv b/packages/ludic.render3d/shaders/spv/grass_cull.comp.spv
new file mode 100644
index 00000000..4b9c9260
Binary files /dev/null and b/packages/ludic.render3d/shaders/spv/grass_cull.comp.spv differ
diff --git a/packages/ludic.render3d/shaders/spv/grass_reset.comp.spv b/packages/ludic.render3d/shaders/spv/grass_reset.comp.spv
new file mode 100644
index 00000000..0d6a3302
Binary files /dev/null and b/packages/ludic.render3d/shaders/spv/grass_reset.comp.spv differ
diff --git a/packages/ludic.render3d/shaders/spv/manifest.txt b/packages/ludic.render3d/shaders/spv/manifest.txt
index c344f908..78000865 100644
--- a/packages/ludic.render3d/shaders/spv/manifest.txt
+++ b/packages/ludic.render3d/shaders/spv/manifest.txt
@@ -843,7 +843,7 @@ T 74258189 u_shadow 6
 T 74258189 u_tershadow 7
 T 74258189 u_ts_height 8
 P d48d6a48 grass.mesh model.frag #define FOLIAGE;#define BLADE;#define MESH;
-B d48d6a48 vert 0 4384
+B d48d6a48 vert 0 4388
 U d48d6a48 vert u_view 0 mat4 1 0
 U d48d6a48 vert u_proj 64 mat4 1 0
 U d48d6a48 vert u_vp 128 mat4 1 0
@@ -862,7 +862,8 @@ U d48d6a48 vert u_tiles 272 vec4 256 16
 U d48d6a48 vert u_s0 4368 float 1 0
 U d48d6a48 vert u_d0 4372 float 1 0
 U d48d6a48 vert u_radius 4376 float 1 0
-U d48d6a48 vert u_dbg 4380 int 1 0
+U d48d6a48 vert u_px 4380 float 1 0
+U d48d6a48 vert u_dbg 4384 int 1 0
 B d48d6a48 frag 1 964
 U d48d6a48 frag u_cascade_vp 0 mat4 5 64
 U d48d6a48 frag u_cascade_split 320 float 5 16
@@ -919,169 +920,171 @@ T d48d6a48 u_prefilter 9
 T d48d6a48 u_shadow 10
 T d48d6a48 u_tershadow 11
 T d48d6a48 u_ts_height 12
-P 7ed7de52 grass.vert model.frag #define FOLIAGE;#define BLADE;
-B 7ed7de52 vert 0 320
-U 7ed7de52 vert u_view 0 mat4 1 0
-U 7ed7de52 vert u_proj 64 mat4 1 0
-U 7ed7de52 vert u_vp 128 mat4 1 0
-U 7ed7de52 vert u_cam_pos 192 vec3 1 0
-U 7ed7de52 vert u_time 204 float 1 0
-U 7ed7de52 vert u_ts_origin 208 vec2 1 0
-U 7ed7de52 vert u_ts_half 216 float 1 0
-U 7ed7de52 vert u_ortho_on 220 float 1 0
-U 7ed7de52 vert u_lake_level 224 float 1 0
-U 7ed7de52 vert u_sea_level 228 float 1 0
-U 7ed7de52 vert u_lake 240 vec4 1 0
-U 7ed7de52 vert u_snow_line 256 float 1 0
-U 7ed7de52 vert u_wind 260 float 1 0
-U 7ed7de52 vert u_push 272 vec3 1 0
-U 7ed7de52 vert u_tile 288 vec2 1 0
-U 7ed7de52 vert u_tile_cells 296 int 1 0
-U 7ed7de52 vert u_per_cell 300 int 1 0
-U 7ed7de52 vert u_s0 304 float 1 0
-U 7ed7de52 vert u_d0 308 float 1 0
-U 7ed7de52 vert u_radius 312 float 1 0
-U 7ed7de52 vert u_dbg 316 int 1 0
-I 7ed7de52 a_pos 0 vec3
-I 7ed7de52 a_uv 2 vec2
-B 7ed7de52 frag 1 964
-U 7ed7de52 frag u_cascade_vp 0 mat4 5 64
-U 7ed7de52 frag u_cascade_split 320 float 5 16
-U 7ed7de52 frag u_cascade_range 400 float 5 16
-U 7ed7de52 frag u_cascade_texel 480 float 5 16
-U 7ed7de52 frag u_sun_dir 560 vec3 1 0
-U 7ed7de52 frag u_sun_color 576 vec3 1 0
-U 7ed7de52 frag u_cam_pos 592 vec3 1 0
-U 7ed7de52 frag u_prefilter_levels 604 float 1 0
-U 7ed7de52 frag u_fog_density 608 float 1 0
-U 7ed7de52 frag u_fog_height_falloff 612 float 1 0
-U 7ed7de52 frag u_fog_base 616 float 1 0
-U 7ed7de52 frag u_clip_y 620 float 1 0
-U 7ed7de52 frag u_spec_scale 624 float 1 0
-U 7ed7de52 frag u_sky_rot 632 vec2 1 0
-U 7ed7de52 frag u_ground_alb 640 vec3 1 0
-U 7ed7de52 frag u_ground_alb_hi 656 vec3 1 0
-U 7ed7de52 frag u_ground_hi_y 668 float 1 0
-U 7ed7de52 frag u_ground_hi_w 672 float 1 0
-U 7ed7de52 frag u_ibl_scale 688 vec3 1 0
-U 7ed7de52 frag u_daylight 700 float 1 0
-U 7ed7de52 frag u_fire_pos 704 vec3 1 0
-U 7ed7de52 frag u_fire_color 720 vec3 1 0
-U 7ed7de52 frag u_hand_pos 736 vec3 1 0
-U 7ed7de52 frag u_hand_color 752 vec3 1 0
-U 7ed7de52 frag u_hand_dir 768 vec3 1 0
-U 7ed7de52 frag u_hand_cone 780 float 1 0
-U 7ed7de52 frag u_hand_reach 784 float 1 0
-U 7ed7de52 frag u_ts_origin 792 vec2 1 0
-U 7ed7de52 frag u_ts_half 800 float 1 0
-U 7ed7de52 frag u_ts_on 804 float 1 0
-U 7ed7de52 frag u_force_cascade 808 int 1 0
-U 7ed7de52 frag u_cloud_shadow 812 float 1 0
-U 7ed7de52 frag u_time 816 float 1 0
-U 7ed7de52 frag u_fog_inscatter 820 float 1 0
-U 7ed7de52 frag u_fog_desat 824 float 1 0
-U 7ed7de52 frag u_model_h 828 float 1 0
-U 7ed7de52 frag u_view 832 mat4 1 0
-U 7ed7de52 frag u_tint 896 vec3 1 0
-U 7ed7de52 frag u_rough_scale 908 float 1 0
-U 7ed7de52 frag u_emissive 912 float 1 0
-U 7ed7de52 frag u_blade_base 928 vec3 1 0
-U 7ed7de52 frag u_blade_tip 944 vec3 1 0
-U 7ed7de52 frag u_blade_cols 956 float 1 0
-U 7ed7de52 frag u_blade_tex_on 960 float 1 0
-T 7ed7de52 u_arm 2
-T 7ed7de52 u_blade_tex 3
-T 7ed7de52 u_brdf 4
-T 7ed7de52 u_diff 5
-T 7ed7de52 u_irradiance 6
-T 7ed7de52 u_nrm 7
-T 7ed7de52 u_ortho 8
-T 7ed7de52 u_prefilter 9
-T 7ed7de52 u_shadow 10
-T 7ed7de52 u_tershadow 11
-T 7ed7de52 u_ts_height 12
-P 7f096322 grass.vert model.frag #define FOLIAGE;#define BLADE;#define TILES;
-B 7f096322 vert 0 4416
-U 7f096322 vert u_view 0 mat4 1 0
-U 7f096322 vert u_proj 64 mat4 1 0
-U 7f096322 vert u_vp 128 mat4 1 0
-U 7f096322 vert u_cam_pos 192 vec3 1 0
-U 7f096322 vert u_time 204 float 1 0
-U 7f096322 vert u_ts_origin 208 vec2 1 0
-U 7f096322 vert u_ts_half 216 float 1 0
-U 7f096322 vert u_ortho_on 220 float 1 0
-U 7f096322 vert u_lake_level 224 float 1 0
-U 7f096322 vert u_sea_level 228 float 1 0
-U 7f096322 vert u_lake 240 vec4 1 0
-U 7f096322 vert u_snow_line 256 float 1 0
-U 7f096322 vert u_wind 260 float 1 0
-U 7f096322 vert u_push 272 vec3 1 0
-U 7f096322 vert u_tile 288 vec2 1 0
-U 7f096322 vert u_tile_cells 296 int 1 0
-U 7f096322 vert u_per_cell 300 int 1 0
-U 7f096322 vert u_tiles 304 vec4 256 16
-U 7f096322 vert u_s0 4400 float 1 0
-U 7f096322 vert u_d0 4404 float 1 0
-U 7f096322 vert u_radius 4408 float 1 0
-U 7f096322 vert u_dbg 4412 int 1 0
-I 7f096322 a_pos 0 vec3
-I 7f096322 a_uv 2 vec2
-B 7f096322 frag 1 964
-U 7f096322 frag u_cascade_vp 0 mat4 5 64
-U 7f096322 frag u_cascade_split 320 float 5 16
-U 7f096322 frag u_cascade_range 400 float 5 16
-U 7f096322 frag u_cascade_texel 480 float 5 16
-U 7f096322 frag u_sun_dir 560 vec3 1 0
-U 7f096322 frag u_sun_color 576 vec3 1 0
-U 7f096322 frag u_cam_pos 592 vec3 1 0
-U 7f096322 frag u_prefilter_levels 604 float 1 0
-U 7f096322 frag u_fog_density 608 float 1 0
-U 7f096322 frag u_fog_height_falloff 612 float 1 0
-U 7f096322 frag u_fog_base 616 float 1 0
-U 7f096322 frag u_clip_y 620 float 1 0
-U 7f096322 frag u_spec_scale 624 float 1 0
-U 7f096322 frag u_sky_rot 632 vec2 1 0
-U 7f096322 frag u_ground_alb 640 vec3 1 0
-U 7f096322 frag u_ground_alb_hi 656 vec3 1 0
-U 7f096322 frag u_ground_hi_y 668 float 1 0
-U 7f096322 frag u_ground_hi_w 672 float 1 0
-U 7f096322 frag u_ibl_scale 688 vec3 1 0
-U 7f096322 frag u_daylight 700 float 1 0
-U 7f096322 frag u_fire_pos 704 vec3 1 0
-U 7f096322 frag u_fire_color 720 vec3 1 0
-U 7f096322 frag u_hand_pos 736 vec3 1 0
-U 7f096322 frag u_hand_color 752 vec3 1 0
-U 7f096322 frag u_hand_dir 768 vec3 1 0
-U 7f096322 frag u_hand_cone 780 float 1 0
-U 7f096322 frag u_hand_reach 784 float 1 0
-U 7f096322 frag u_ts_origin 792 vec2 1 0
-U 7f096322 frag u_ts_half 800 float 1 0
-U 7f096322 frag u_ts_on 804 float 1 0
-U 7f096322 frag u_force_cascade 808 int 1 0
-U 7f096322 frag u_cloud_shadow 812 float 1 0
-U 7f096322 frag u_time 816 float 1 0
-U 7f096322 frag u_fog_inscatter 820 float 1 0
-U 7f096322 frag u_fog_desat 824 float 1 0
-U 7f096322 frag u_model_h 828 float 1 0
-U 7f096322 frag u_view 832 mat4 1 0
-U 7f096322 frag u_tint 896 vec3 1 0
-U 7f096322 frag u_rough_scale 908 float 1 0
-U 7f096322 frag u_emissive 912 float 1 0
-U 7f096322 frag u_blade_base 928 vec3 1 0
-U 7f096322 frag u_blade_tip 944 vec3 1 0
-U 7f096322 frag u_blade_cols 956 float 1 0
-U 7f096322 frag u_blade_tex_on 960 float 1 0
-T 7f096322 u_arm 2
-T 7f096322 u_blade_tex 3
-T 7f096322 u_brdf 4
-T 7f096322 u_diff 5
-T 7f096322 u_irradiance 6
-T 7f096322 u_nrm 7
-T 7f096322 u_ortho 8
-T 7f096322 u_prefilter 9
-T 7f096322 u_shadow 10
-T 7f096322 u_tershadow 11
-T 7f096322 u_ts_height 12
+P cb13f3c6 grass.vert model.frag #define FOLIAGE;#define BLADE;#define GBLADE;
+B cb13f3c6 vert 0 324
+U cb13f3c6 vert u_view 0 mat4 1 0
+U cb13f3c6 vert u_proj 64 mat4 1 0
+U cb13f3c6 vert u_vp 128 mat4 1 0
+U cb13f3c6 vert u_cam_pos 192 vec3 1 0
+U cb13f3c6 vert u_time 204 float 1 0
+U cb13f3c6 vert u_ts_origin 208 vec2 1 0
+U cb13f3c6 vert u_ts_half 216 float 1 0
+U cb13f3c6 vert u_ortho_on 220 float 1 0
+U cb13f3c6 vert u_lake_level 224 float 1 0
+U cb13f3c6 vert u_sea_level 228 float 1 0
+U cb13f3c6 vert u_lake 240 vec4 1 0
+U cb13f3c6 vert u_snow_line 256 float 1 0
+U cb13f3c6 vert u_wind 260 float 1 0
+U cb13f3c6 vert u_push 272 vec3 1 0
+U cb13f3c6 vert u_tile 288 vec2 1 0
+U cb13f3c6 vert u_tile_cells 296 int 1 0
+U cb13f3c6 vert u_per_cell 300 int 1 0
+U cb13f3c6 vert u_s0 304 float 1 0
+U cb13f3c6 vert u_d0 308 float 1 0
+U cb13f3c6 vert u_radius 312 float 1 0
+U cb13f3c6 vert u_px 316 float 1 0
+U cb13f3c6 vert u_dbg 320 int 1 0
+I cb13f3c6 a_pos 0 vec3
+I cb13f3c6 a_uv 2 vec2
+B cb13f3c6 frag 1 964
+U cb13f3c6 frag u_cascade_vp 0 mat4 5 64
+U cb13f3c6 frag u_cascade_split 320 float 5 16
+U cb13f3c6 frag u_cascade_range 400 float 5 16
+U cb13f3c6 frag u_cascade_texel 480 float 5 16
+U cb13f3c6 frag u_sun_dir 560 vec3 1 0
+U cb13f3c6 frag u_sun_color 576 vec3 1 0
+U cb13f3c6 frag u_cam_pos 592 vec3 1 0
+U cb13f3c6 frag u_prefilter_levels 604 float 1 0
+U cb13f3c6 frag u_fog_density 608 float 1 0
+U cb13f3c6 frag u_fog_height_falloff 612 float 1 0
+U cb13f3c6 frag u_fog_base 616 float 1 0
+U cb13f3c6 frag u_clip_y 620 float 1 0
+U cb13f3c6 frag u_spec_scale 624 float 1 0
+U cb13f3c6 frag u_sky_rot 632 vec2 1 0
+U cb13f3c6 frag u_ground_alb 640 vec3 1 0
+U cb13f3c6 frag u_ground_alb_hi 656 vec3 1 0
+U cb13f3c6 frag u_ground_hi_y 668 float 1 0
+U cb13f3c6 frag u_ground_hi_w 672 float 1 0
+U cb13f3c6 frag u_ibl_scale 688 vec3 1 0
+U cb13f3c6 frag u_daylight 700 float 1 0
+U cb13f3c6 frag u_fire_pos 704 vec3 1 0
+U cb13f3c6 frag u_fire_color 720 vec3 1 0
+U cb13f3c6 frag u_hand_pos 736 vec3 1 0
+U cb13f3c6 frag u_hand_color 752 vec3 1 0
+U cb13f3c6 frag u_hand_dir 768 vec3 1 0
+U cb13f3c6 frag u_hand_cone 780 float 1 0
+U cb13f3c6 frag u_hand_reach 784 float 1 0
+U cb13f3c6 frag u_ts_origin 792 vec2 1 0
+U cb13f3c6 frag u_ts_half 800 float 1 0
+U cb13f3c6 frag u_ts_on 804 float 1 0
+U cb13f3c6 frag u_force_cascade 808 int 1 0
+U cb13f3c6 frag u_cloud_shadow 812 float 1 0
+U cb13f3c6 frag u_time 816 float 1 0
+U cb13f3c6 frag u_fog_inscatter 820 float 1 0
+U cb13f3c6 frag u_fog_desat 824 float 1 0
+U cb13f3c6 frag u_model_h 828 float 1 0
+U cb13f3c6 frag u_view 832 mat4 1 0
+U cb13f3c6 frag u_tint 896 vec3 1 0
+U cb13f3c6 frag u_rough_scale 908 float 1 0
+U cb13f3c6 frag u_emissive 912 float 1 0
+U cb13f3c6 frag u_blade_base 928 vec3 1 0
+U cb13f3c6 frag u_blade_tip 944 vec3 1 0
+U cb13f3c6 frag u_blade_cols 956 float 1 0
+U cb13f3c6 frag u_blade_tex_on 960 float 1 0
+T cb13f3c6 u_arm 2
+T cb13f3c6 u_blade_tex 3
+T cb13f3c6 u_brdf 4
+T cb13f3c6 u_diff 5
+T cb13f3c6 u_irradiance 6
+T cb13f3c6 u_nrm 7
+T cb13f3c6 u_ortho 8
+T cb13f3c6 u_prefilter 9
+T cb13f3c6 u_shadow 10
+T cb13f3c6 u_tershadow 11
+T cb13f3c6 u_ts_height 12
+P b8cff226 grass.vert model.frag #define FOLIAGE;#define BLADE;#define GBLADE;#define TILES;
+B b8cff226 vert 0 4420
+U b8cff226 vert u_view 0 mat4 1 0
+U b8cff226 vert u_proj 64 mat4 1 0
+U b8cff226 vert u_vp 128 mat4 1 0
+U b8cff226 vert u_cam_pos 192 vec3 1 0
+U b8cff226 vert u_time 204 float 1 0
+U b8cff226 vert u_ts_origin 208 vec2 1 0
+U b8cff226 vert u_ts_half 216 float 1 0
+U b8cff226 vert u_ortho_on 220 float 1 0
+U b8cff226 vert u_lake_level 224 float 1 0
+U b8cff226 vert u_sea_level 228 float 1 0
+U b8cff226 vert u_lake 240 vec4 1 0
+U b8cff226 vert u_snow_line 256 float 1 0
+U b8cff226 vert u_wind 260 float 1 0
+U b8cff226 vert u_push 272 vec3 1 0
+U b8cff226 vert u_tile 288 vec2 1 0
+U b8cff226 vert u_tile_cells 296 int 1 0
+U b8cff226 vert u_per_cell 300 int 1 0
+U b8cff226 vert u_tiles 304 vec4 256 16
+U b8cff226 vert u_s0 4400 float 1 0
+U b8cff226 vert u_d0 4404 float 1 0
+U b8cff226 vert u_radius 4408 float 1 0
+U b8cff226 vert u_px 4412 float 1 0
+U b8cff226 vert u_dbg 4416 int 1 0
+I b8cff226 a_pos 0 vec3
+I b8cff226 a_uv 2 vec2
+B b8cff226 frag 1 964
+U b8cff226 frag u_cascade_vp 0 mat4 5 64
+U b8cff226 frag u_cascade_split 320 float 5 16
+U b8cff226 frag u_cascade_range 400 float 5 16
+U b8cff226 frag u_cascade_texel 480 float 5 16
+U b8cff226 frag u_sun_dir 560 vec3 1 0
+U b8cff226 frag u_sun_color 576 vec3 1 0
+U b8cff226 frag u_cam_pos 592 vec3 1 0
+U b8cff226 frag u_prefilter_levels 604 float 1 0
+U b8cff226 frag u_fog_density 608 float 1 0
+U b8cff226 frag u_fog_height_falloff 612 float 1 0
+U b8cff226 frag u_fog_base 616 float 1 0
+U b8cff226 frag u_clip_y 620 float 1 0
+U b8cff226 frag u_spec_scale 624 float 1 0
+U b8cff226 frag u_sky_rot 632 vec2 1 0
+U b8cff226 frag u_ground_alb 640 vec3 1 0
+U b8cff226 frag u_ground_alb_hi 656 vec3 1 0
+U b8cff226 frag u_ground_hi_y 668 float 1 0
+U b8cff226 frag u_ground_hi_w 672 float 1 0
+U b8cff226 frag u_ibl_scale 688 vec3 1 0
+U b8cff226 frag u_daylight 700 float 1 0
+U b8cff226 frag u_fire_pos 704 vec3 1 0
+U b8cff226 frag u_fire_color 720 vec3 1 0
+U b8cff226 frag u_hand_pos 736 vec3 1 0
+U b8cff226 frag u_hand_color 752 vec3 1 0
+U b8cff226 frag u_hand_dir 768 vec3 1 0
+U b8cff226 frag u_hand_cone 780 float 1 0
+U b8cff226 frag u_hand_reach 784 float 1 0
+U b8cff226 frag u_ts_origin 792 vec2 1 0
+U b8cff226 frag u_ts_half 800 float 1 0
+U b8cff226 frag u_ts_on 804 float 1 0
+U b8cff226 frag u_force_cascade 808 int 1 0
+U b8cff226 frag u_cloud_shadow 812 float 1 0
+U b8cff226 frag u_time 816 float 1 0
+U b8cff226 frag u_fog_inscatter 820 float 1 0
+U b8cff226 frag u_fog_desat 824 float 1 0
+U b8cff226 frag u_model_h 828 float 1 0
+U b8cff226 frag u_view 832 mat4 1 0
+U b8cff226 frag u_tint 896 vec3 1 0
+U b8cff226 frag u_rough_scale 908 float 1 0
+U b8cff226 frag u_emissive 912 float 1 0
+U b8cff226 frag u_blade_base 928 vec3 1 0
+U b8cff226 frag u_blade_tip 944 vec3 1 0
+U b8cff226 frag u_blade_cols 956 float 1 0
+U b8cff226 frag u_blade_tex_on 960 float 1 0
+T b8cff226 u_arm 2
+T b8cff226 u_blade_tex 3
+T b8cff226 u_brdf 4
+T b8cff226 u_diff 5
+T b8cff226 u_irradiance 6
+T b8cff226 u_nrm 7
+T b8cff226 u_ortho 8
+T b8cff226 u_prefilter 9
+T b8cff226 u_shadow 10
+T b8cff226 u_tershadow 11
+T b8cff226 u_ts_height 12
 P 3733fc38 impostor.vert impostor.frag 
 B 3733fc38 vert 0 232
 U 3733fc38 vert u_view 0 mat4 1 0
@@ -3522,3 +3525,72 @@ T 46e2325c u_scene 7
 T 46e2325c u_shadow 8
 T 46e2325c u_tershadow 9
 T 46e2325c u_ts_height 10
+P b20fdcb5 grass_inst.vert model.frag #define FOLIAGE;#define BLADE;#define GBLADE;#define GINST;
+B b20fdcb5 vert 0 160
+U b20fdcb5 vert u_view 0 mat4 1 0
+U b20fdcb5 vert u_proj 64 mat4 1 0
+U b20fdcb5 vert u_time 128 float 1 0
+U b20fdcb5 vert u_wind 132 float 1 0
+U b20fdcb5 vert u_push 144 vec3 1 0
+U b20fdcb5 vert u_dbg 156 int 1 0
+I b20fdcb5 i_root 3 vec4
+I b20fdcb5 i_shape 4 vec4
+I b20fdcb5 i_ground 6 vec4
+I b20fdcb5 a_pos 0 vec3
+I b20fdcb5 i_tint 5 vec4
+I b20fdcb5 a_uv 2 vec2
+B b20fdcb5 frag 1 964
+U b20fdcb5 frag u_cascade_vp 0 mat4 5 64
+U b20fdcb5 frag u_cascade_split 320 float 5 16
+U b20fdcb5 frag u_cascade_range 400 float 5 16
+U b20fdcb5 frag u_cascade_texel 480 float 5 16
+U b20fdcb5 frag u_sun_dir 560 vec3 1 0
+U b20fdcb5 frag u_sun_color 576 vec3 1 0
+U b20fdcb5 frag u_cam_pos 592 vec3 1 0
+U b20fdcb5 frag u_prefilter_levels 604 float 1 0
+U b20fdcb5 frag u_fog_density 608 float 1 0
+U b20fdcb5 frag u_fog_height_falloff 612 float 1 0
+U b20fdcb5 frag u_fog_base 616 float 1 0
+U b20fdcb5 frag u_clip_y 620 float 1 0
+U b20fdcb5 frag u_spec_scale 624 float 1 0
+U b20fdcb5 frag u_sky_rot 632 vec2 1 0
+U b20fdcb5 frag u_ground_alb 640 vec3 1 0
+U b20fdcb5 frag u_ground_alb_hi 656 vec3 1 0
+U b20fdcb5 frag u_ground_hi_y 668 float 1 0
+U b20fdcb5 frag u_ground_hi_w 672 float 1 0
+U b20fdcb5 frag u_ibl_scale 688 vec3 1 0
+U b20fdcb5 frag u_daylight 700 float 1 0
+U b20fdcb5 frag u_fire_pos 704 vec3 1 0
+U b20fdcb5 frag u_fire_color 720 vec3 1 0
+U b20fdcb5 frag u_hand_pos 736 vec3 1 0
+U b20fdcb5 frag u_hand_color 752 vec3 1 0
+U b20fdcb5 frag u_hand_dir 768 vec3 1 0
+U b20fdcb5 frag u_hand_cone 780 float 1 0
+U b20fdcb5 frag u_hand_reach 784 float 1 0
+U b20fdcb5 frag u_ts_origin 792 vec2 1 0
+U b20fdcb5 frag u_ts_half 800 float 1 0
+U b20fdcb5 frag u_ts_on 804 float 1 0
+U b20fdcb5 frag u_force_cascade 808 int 1 0
+U b20fdcb5 frag u_cloud_shadow 812 float 1 0
+U b20fdcb5 frag u_time 816 float 1 0
+U b20fdcb5 frag u_fog_inscatter 820 float 1 0
+U b20fdcb5 frag u_fog_desat 824 float 1 0
+U b20fdcb5 frag u_model_h 828 float 1 0
+U b20fdcb5 frag u_view 832 mat4 1 0
+U b20fdcb5 frag u_tint 896 vec3 1 0
+U b20fdcb5 frag u_rough_scale 908 float 1 0
+U b20fdcb5 frag u_emissive 912 float 1 0
+U b20fdcb5 frag u_blade_base 928 vec3 1 0
+U b20fdcb5 frag u_blade_tip 944 vec3 1 0
+U b20fdcb5 frag u_blade_cols 956 float 1 0
+U b20fdcb5 frag u_blade_tex_on 960 float 1 0
+T b20fdcb5 u_arm 2
+T b20fdcb5 u_blade_tex 3
+T b20fdcb5 u_brdf 4
+T b20fdcb5 u_diff 5
+T b20fdcb5 u_irradiance 6
+T b20fdcb5 u_nrm 7
+T b20fdcb5 u_prefilter 8
+T b20fdcb5 u_shadow 9
+T b20fdcb5 u_tershadow 10
+T b20fdcb5 u_ts_height 11
diff --git a/packages/ludic.render3d/shaders/variants.list b/packages/ludic.render3d/shaders/variants.list
index fee8e402..340d0ad3 100644
--- a/packages/ludic.render3d/shaders/variants.list
+++ b/packages/ludic.render3d/shaders/variants.list
@@ -20,8 +20,8 @@ fullscreen.vert|tonemap.frag|
 fullscreen.vert|tonemap.frag|#define HDR10;
 fullscreen.vert|volumetric.frag|
 grass.mesh|model.frag|#define FOLIAGE;#define BLADE;#define MESH;
-grass.vert|model.frag|#define FOLIAGE;#define BLADE;
-grass.vert|model.frag|#define FOLIAGE;#define BLADE;#define TILES;
+grass.vert|model.frag|#define FOLIAGE;#define BLADE;#define GBLADE;
+grass.vert|model.frag|#define FOLIAGE;#define BLADE;#define GBLADE;#define TILES;
 impostor.vert|impostor.frag|
 impostor.vert|impostor.frag|#define SHADOW_PASS;
 model.vert|bake.frag|
@@ -59,3 +59,4 @@ terrain.vert|terrain.frag|#define SUN_INLINE;#define FAR_ONLY;
 terrain.vert|terrain.frag|#define SUN_INLINE;#define NEAR_ONLY;
 terrain.vert|tersun.frag|
 water.vert|water.frag|
+grass_inst.vert|model.frag|#define FOLIAGE;#define BLADE;#define GBLADE;#define GINST;
diff --git a/tools/ludic-cli/shaders.ludic b/tools/ludic-cli/shaders.ludic
index 16389dce..a27cf2ff 100644
--- a/tools/ludic-cli/shaders.ludic
+++ b/tools/ludic-cli/shaders.ludic
@@ -234,7 +234,10 @@ function cmd_shaders() -> int {
     let depth_remap = "\n#undef main\nvoid main() {\n  ludic_gl_main();\n  gl_Position.z = (gl_Position.z + gl_Position.w) * 0.5;\n}\n"
     # invariant: the foliage's depth prepass and its lit pass (depth EQUAL) are different variants,
     # and without it the compiler may compute their positions a hair apart and fail the test
-    var vsrc = "#version 460\n" + defines + shd_wind + "invariant gl_Position;\n#define main ludic_gl_main\n" + shd_file(parts[0]) + depth_remap
+    # as programs.ludic: a blade's vertex stage gets the noise its colour field needs
+    var vnoise = ""
+    if Text.contains(defines, "#define GBLADE") { vnoise = shd_noise }
+    var vsrc = "#version 460\n" + defines + shd_wind + vnoise + "invariant gl_Position;\n#define main ludic_gl_main\n" + shd_file(parts[0]) + depth_remap
     # A mesh shader stands in for the vertex stage (a variant whose first file is *.mesh): it writes an
     # array of positions, so it cannot be wrapped for the depth remap or take invariant gl_Position -
     # it remaps depth itself. Its SPIR-V keeps the .vert name, so the manifest and loader are unchanged.
@@ -304,7 +307,8 @@ function cmd_shaders() -> int {
       let cp = Text.split(cl, "|")
       if len(cp) < 2 { err(`shaders: bad compute line: {cl}\n`);  return 1 }
       let glsl = `{tmp}/{cp[0]}.comp.glsl`
-      if not write_file(glsl, "#version 460\n" + shd_file(cp[1])) { err(`shaders: cannot write {glsl}\n`);  return 1 }
+      # the noise functions come first, as in a fragment stage (grass_cull uses them; the rest ignore them)
+      if not write_file(glsl, "#version 460\n" + shd_noise + shd_file(cp[1])) { err(`shaders: cannot write {glsl}\n`);  return 1 }
       let spv = `{outdir}/{cp[0]}.comp.spv`
       let logf = `{tmp}/{cp[0]}.comp.log`
       if not shq(`{shd_q(shd_bin + "/glslangValidator")} -V -S comp {shd_q(glsl)} -o {shd_q(spv)} > {shd_q(logf)} 2>&1`) {