diff --git a/README.md b/README.md
index 05119d3..b575e55 100644
--- a/README.md
+++ b/README.md
@@ -1,5 +1,11 @@
 # multicam-sim
 
+![Eight ring cameras render one moving object from eight angles; a red identity box tracks the same object across every view that sees it.](assets/multiview_hero.gif)
+
+*multi-view: same object, 8 angles, one identity — eight ring cameras rendered by
+the pure-numpy rasterizer, with the object's analytic projected box drawn in every
+view that sees it. Reproduce: `uv run --with pillow python scripts/render_multiview_hero.py`.*
+
 ![Three MTMC camera stations with non-overlapping fields of view watch one object cross a corridor on a synced timeline; it hands off from camera 0 to 1 to 2, with blind-gap frames where no camera sees it.](docs/assets/hero_grid.gif)
 
 *One object, three cameras with **disjoint** views, one timeline — a green border
diff --git a/assets/multiview_hero.gif b/assets/multiview_hero.gif
new file mode 100644
index 0000000..a7cfb0e
Binary files /dev/null and b/assets/multiview_hero.gif differ
diff --git a/scripts/render_multiview_hero.py b/scripts/render_multiview_hero.py
new file mode 100644
index 0000000..f8d18a1
--- /dev/null
+++ b/scripts/render_multiview_hero.py
@@ -0,0 +1,124 @@
+"""Render the multi-view hero GIF: one moving object, 8 angles, one identity.
+
+Deterministic and dependency-light: the pure-numpy :class:`RasterizerBackend`
+renders each of 8 ring cameras every frame, the analytic manifest gives the
+object's projected pixel in every view, and we overlay the SAME id + colour in
+all 8 tiles so a reader can see it is one object seen from eight angles. Tiled
+2x4 with Pillow into ``assets/multiview_hero.gif``.
+
+``Pillow`` is imported lazily inside :func:`main`, never at module load, so the
+package/test import path (and CI) never depends on it. Run:
+
+    uv run --with pillow python scripts/render_multiview_hero.py
+"""
+
+from __future__ import annotations
+
+from pathlib import Path as FsPath
+
+import numpy as np
+
+from multicam_sim.dsl import CameraRig, Path, SceneBuilder
+from multicam_sim.dsl.raster import RasterizerBackend, RasterizerConfig
+from multicam_sim.manifest import build_manifest
+
+# --- fixed, reproducible parameters ---------------------------------------- #
+N_CAMS = 8
+RADIUS = 4.0
+HEIGHT = 1.5
+LOOK_AT = (0.0, 0.0, 0.5)
+FOCAL = 500.0
+RES = 256
+NUM_FRAMES = 40
+FPS = 30.0
+ENTITY_ID = "e0"
+POINT_NAME = "center"
+PATH_START = (-1.2, -0.5, 0.5)
+PATH_END = (1.2, 0.5, 0.9)
+
+TILE = 128  # per-camera tile size in the grid (renders downscaled from RES)
+GRID_COLS, GRID_ROWS = 4, 2  # 2x4 = 8 cameras
+OBJ_RADIUS = 0.12  # world-space object radius (bolder, readable sphere)
+BOX_HALF = 22  # half-size (px, at RES) of the identity box drawn on the object
+OBJ_COLOR = (255, 90, 70)  # one colour == one identity across all 8 views
+GIF_MS = 90
+
+OUT = FsPath(__file__).resolve().parents[1] / "assets" / "multiview_hero.gif"
+
+
+def build_scene():
+    cams = CameraRig.ring(
+        n=N_CAMS,
+        radius=RADIUS,
+        height=HEIGHT,
+        look_at=LOOK_AT,
+        width=RES,
+        height_px=RES,
+        focal=FOCAL,
+    )
+    return (
+        SceneBuilder(fps=FPS, num_frames=NUM_FRAMES)
+        .cameras(cams)
+        .entity(ENTITY_ID, Path.linear(PATH_START, PATH_END))
+        .build()
+    )
+
+
+def main() -> None:
+    import time
+
+    from PIL import Image, ImageDraw  # lazy: never imported at package/test load
+
+    t0 = time.time()
+    scene = build_scene()
+    manifest = build_manifest(scene)
+    backend = RasterizerBackend(RasterizerConfig(point_radius=OBJ_RADIUS))
+
+    # per (frame, cam): the object's projected uv + whether it is in view.
+    obs: dict[tuple[int, int], tuple[float, float, bool]] = {}
+    entity = next(e for e in manifest.entities if e.id == ENTITY_ID)
+    for fr in entity.frames:
+        for pc in fr.points[POINT_NAME].per_cam:
+            obs[(fr.frame, pc.cam)] = (pc.uv[0], pc.uv[1], bool(pc.in_view))
+
+    grid_w, grid_h = GRID_COLS * TILE, GRID_ROWS * TILE
+    frames: list[Image.Image] = []
+    for f in range(NUM_FRAMES):
+        grid = Image.new("RGB", (grid_w, grid_h), (18, 18, 22))
+        for cam_id in range(N_CAMS):
+            img = np.asarray(backend.render(scene, cam_id, f), dtype=np.uint8)
+            tile = Image.fromarray(img, "RGB")
+            draw = ImageDraw.Draw(tile)
+            u, v, in_view = obs[(f, cam_id)]
+            if in_view:
+                draw.rectangle(
+                    [u - BOX_HALF, v - BOX_HALF, u + BOX_HALF, v + BOX_HALF],
+                    outline=OBJ_COLOR,
+                    width=3,
+                )
+                draw.text((u - BOX_HALF, v - BOX_HALF - 12), ENTITY_ID, fill=OBJ_COLOR)
+            draw.text((4, 4), f"cam {cam_id}", fill=(210, 210, 210))
+            tile = tile.resize((TILE, TILE), Image.BILINEAR)
+            col, row = cam_id % GRID_COLS, cam_id // GRID_COLS
+            grid.paste(tile, (col * TILE, row * TILE))
+        frames.append(grid)
+
+    OUT.parent.mkdir(parents=True, exist_ok=True)
+    frames[0].save(
+        OUT,
+        save_all=True,
+        append_images=frames[1:],
+        duration=GIF_MS,
+        loop=0,
+        optimize=True,
+    )
+    size_kb = OUT.stat().st_size / 1024
+    print(
+        f"wrote {OUT.relative_to(FsPath.cwd())}  "
+        f"{NUM_FRAMES} frames  {grid_w}x{grid_h}  {size_kb:.1f} KiB  "
+        f"{time.time() - t0:.1f}s"
+    )
+
+
+if __name__ == "__main__":
+    main()