diff --git a/changes/lab-png-16-bit.md b/changes/lab-png-16-bit.md
new file mode 100644
index 00000000..41af6d5a
--- /dev/null
+++ b/changes/lab-png-16-bit.md
@@ -0,0 +1,3 @@
+bump: patch
+type: feature
+**ludic.lab: 16-bit PNGs** — `lab_png_write16_from(st, path, w, h, channels, px, at)` writes 16-bit samples (two bytes each, most significant first) the way `lab_png_write_from` writes 8-bit ones, so a test can write a height map render3d's decoder reads back as R16; `lp_png_bytes(stride, h)` is the size of one.
diff --git a/packages/ludic.lab/README.md b/packages/ludic.lab/README.md
index 34ca8069..c7c7ae22 100644
--- a/packages/ludic.lab/README.md
+++ b/packages/ludic.lab/README.md
@@ -61,7 +61,7 @@ showing its own models - adds them from its own module with `def LabScenes <name
 | `lab_settle(frames)`, `lab_font(dir)`, `lab_title(s)`, `lab_note(s)` | how long a shot settles; the overlay font for captions (none by default); the window's title; a line under the caption |
 | `lab_scene()`, `lab_frame()`, `lab_time()` | the scene, and the lab's clock |
 | `lab_ppm_to_png(ppm, png)` | a frame as a PNG (stored, not compressed) |
-| `lab_png_write(path, w, h, channels, px)`, `lab_png_write_from(..., px, at)` | 8-bit pixels (1 grey, 2 grey + alpha, 3 RGB, 4 RGBA) as a PNG, the same way; `png_write.ludic` needs nothing else of the lab, so a program with a lab of its own imports it alone (`import "ludic.lab/png_write.ludic"`, state `LabPngState`) |
+| `lab_png_write(path, w, h, channels, px)`, `lab_png_write_from(..., px, at)`, `lab_png_write16_from(..., px, at)` | 8-bit pixels (1 grey, 2 grey + alpha, 3 RGB, 4 RGBA) as a PNG, the same way, or 16-bit ones (two bytes a sample, most significant first: a height map a test writes); `png_write.ludic` needs nothing else of the lab, so a program with a lab of its own imports it alone (`import "ludic.lab/png_write.ludic"`, state `LabPngState`) |
 | `lab_px_r32f_norm`, `lab_px_rg16f`, `lab_px_rgba16f_hdr` (`png_convert.ludic`) | previews of float textures as 8-bit pixels: R32F min..max to black..white; RG16F as red and green x255, blue 0; HDR RGBA16F as x/(1+x) then sRGB, alpha dropped |
 
 ## The plate's files
diff --git a/packages/ludic.lab/png_write.ludic b/packages/ludic.lab/png_write.ludic
index af01ba8e..4cd24620 100644
--- a/packages/ludic.lab/png_write.ludic
+++ b/packages/ludic.lab/png_write.ludic
@@ -1,4 +1,4 @@
-# png_write.ludic - raw 8-bit pixels (1 to 4 channels) as a PNG anything opens: filter 0 on every row,
+# png_write.ludic - raw 8- or 16-bit pixels (1 to 4 channels) as a PNG anything opens: filter 0 on every row,
 # deflate's stored blocks (no zlib here), the CRCs and the Adler-32. Needs nothing else of the lab, so a
 # program with a lab of its own imports this file alone: import "ludic.lab/png_write.ludic"
 
@@ -48,28 +48,30 @@ function lp_colour_type(channels: int) -> int {
 
 # the header, then the pixels, then the end, into one buffer; the length written, or -1
 @alloc_ok("a bake's preview or a lab shot: one image written, once")
-function lp_encode(table: []long, w: int, h: int, channels: int, px: []byte, at: int, out: []byte) -> int {
+function lp_encode(table: []long, w: int, h: int, channels: int, px: []byte, at: int, out: []byte) -> int { return lp_encode_depth(table, w, h, channels, 8, px, at, out) }
+# ... with `depth` bits a sample (8, or 16: two bytes each, most significant first, as PNG stores them)
+@alloc_ok("a bake's preview or a lab shot: one image written, once")
+function lp_encode_depth(table: []long, w: int, h: int, channels: int, depth: int, px: []byte, at: int, out: []byte) -> int {
   let sig = [137, 80, 78, 71, 13, 10, 26, 10]
   for i in 0 .. 8 { out[i] = sig[i] }
   lp_put32(out, 8, long(13))
   lp_tag(out, 12, "IHDR")
   lp_put32(out, 16, long(w))
   lp_put32(out, 20, long(h))
-  out[24] = 8
+  out[24] = depth
   out[25] = lp_colour_type(channels)
   out[26] = 0
   out[27] = 0
   out[28] = 0
   lp_put32(out, 29, lp_crc_of(table, out, 12, 17))
-  let o = lp_idat(table, w, h, channels, px, at, out, 33)
+  let o = lp_idat(table, w * channels * depth / 8, h, px, at, out, 33)
   lp_put32(out, o, long(0))
   lp_tag(out, o + 4, "IEND")
   lp_put32(out, o + 8, lp_crc_of(table, out, o + 4, 4))
   return o + 12
 }
-# the IDAT chunk at `at0`: every row behind a filter byte of 0, in stored blocks; where it ends
-function lp_idat(table: []long, w: int, h: int, channels: int, px: []byte, at: int, out: []byte, at0: int) -> int {
-  let stride = w * channels
+# the IDAT chunk at `at0`: every row (`stride` bytes) behind a filter byte of 0, in stored blocks; where it ends
+function lp_idat(table: []long, stride: int, h: int, px: []byte, at: int, out: []byte, at0: int) -> int {
   let row = stride + 1
   let raw = row * h
   let blocks = (raw + 65534) / 65535
@@ -112,8 +114,10 @@ function lp_idat(table: []long, w: int, h: int, channels: int, px: []byte, at: i
 }
 
 # the bytes a PNG of this size takes, stored
-function lp_png_size(w: int, h: int, channels: int) -> int {
-  let raw = (w * channels + 1) * h
+function lp_png_size(w: int, h: int, channels: int) -> int { return lp_png_bytes(w * channels, h) }
+# ... of rows `stride` bytes long
+function lp_png_bytes(stride: int, h: int) -> int {
+  let raw = (stride + 1) * h
   return 8 + 25 + 12 + 2 + ((raw + 65534) / 65535) * 5 + raw + 4 + 12
 }
 
@@ -131,3 +135,14 @@ export function lab_png_write_from(lab_png_st: LabPngState, path: string, w: int
 export function lab_png_write(lab_png_st: LabPngState, path: string, w: int, h: int, channels: int, px: []byte) -> bool {
   return lab_png_write_from(lab_png_st, path, w, h, channels, px, 0)
 }
+# w x h pixels of `channels` 16-bit samples each, two bytes a sample most significant first (a height map:
+# 1 channel, read back by render3d's decoder as R16), rows top down, starting at px[at]
+@alloc_ok("a bake or a test's fixture: one image written, once")
+export function lab_png_write16_from(lab_png_st: LabPngState, path: string, w: int, h: int, channels: int, px: []byte, at: int) -> bool {
+  if channels < 1 or channels > 4 or w <= 0 or h <= 0 or at + w * h * channels * 2 > len(px) { return false }
+  let out = buffer(lp_png_bytes(w * channels * 2, h))
+  let n = lp_encode_depth(lab_png_st.table, w, h, channels, 16, px, at, out)
+  let ok = Fs.write_bytes(path, out, n)
+  free(out)
+  return ok
+}
diff --git a/packages/ludic.lab/tests/lab_test.ludic b/packages/ludic.lab/tests/lab_test.ludic
index ffa25792..23e01155 100644
--- a/packages/ludic.lab/tests/lab_test.ludic
+++ b/packages/ludic.lab/tests/lab_test.ludic
@@ -47,6 +47,18 @@ program LabTest {
     expect(not lab_png_write(lab_png_st, `{dir}/lab_test_short.png`, 9, 9, 4, px))
   }
 
+  test "16-bit samples become a 16-bit PNG, two bytes a sample" (lab_png_st: LabPngState) {
+    let dir = Os.temp_dir()
+    let px = buffer(3 * 2 * 2)
+    for i in 0 .. 12 { px[i] = i * 20 }
+    expect(lab_png_write16_from(lab_png_st, `{dir}/lab_test_r16.png`, 3, 2, 1, px, 0))
+    let p = Fs.read_bytes(`{dir}/lab_test_r16.png`)
+    expect_eq(p[24], 16)                       # bit depth
+    expect_eq(p[25], 0)                        # colour type grey
+    expect_eq(len(p), lp_png_bytes(3 * 2, 2))
+    expect(not lab_png_write16_from(lab_png_st, `{dir}/lab_test_r16_short.png`, 3, 3, 1, px, 0))
+  }
+
   test "a half float and a float read as the numbers they store" {
     let b = buffer(8)
     b[0] = 0;  b[1] = 60                       # half 1.0 = 0x3C00