Strengthen strict overhead sprite prompting

app.py

@@ -1947,6 +1947,61 @@ def neural_foreground_cutout(image: Image.Image) -> Image.Image:
     return cutout
 
 
+def top_down_subject_contract(spec: AssetSpec) -> str:
+    """Describe visible overhead anatomy without drawing or reshaping model artwork."""
+    # Classify only from the user's visual description. Custom IDs such as
+    # ``player`` or ``enemy`` are integration hooks, not reliable anatomy.
+    description = f"{spec.description} {spec.prompt}".lower()
+    humanoid_terms = (
+        "humanoid",
+        "character",
+        "player",
+        "person",
+        "human",
+        "astronaut",
+        "soldier",
+        "warrior",
+        "ranger",
+    )
+    radial_terms = (
+        "radial",
+        "six-legged",
+        "six legged",
+        "multi-legged",
+        "multi legged",
+        "spider",
+        "crab",
+        "quadruped",
+    )
+    vehicle_terms = (
+        "vehicle",
+        "rover",
+        "car",
+        "truck",
+        "tank",
+        "motorcycle",
+        "spaceship",
+        "spacecraft",
+    )
+
+    if contains_any_term(description, radial_terms):
+        return (
+            "Top of body dominates; splayed limbs radiate in image plane; "
+            "no upright torso or side elevation."
+        )
+    if contains_any_term(description, vehicle_terms):
+        return (
+            "Roof and chassis top dominate; wheel footprints flank both sides; "
+            "no grille or side elevation."
+        )
+    if spec.silhouette == "humanoid" or contains_any_term(description, humanoid_terms):
+        return (
+            "Full body overhead: head crown, shoulders and back visible; "
+            "face and chest invisible; limbs foreshortened."
+        )
+    return "Flattened overhead footprint with top surfaces dominant; no upright or eye-level presentation."
+
+
 def primary_diffusion_prompt(spec: AssetSpec) -> str:
     camera = primary_camera_phrase(spec.camera, is_background_spec(spec))
     if spec.camera == "auto" and spec.camera_instruction and spec.camera_instruction.lower() != "auto":
@@ -1957,12 +2012,11 @@ def primary_diffusion_prompt(spec: AssetSpec) -> str:
         variation = f"Variation: {compact_prompt_words(spec.variation, 4)}. " if spec.variation else ""
         subject = (spec.group or spec.role).replace("_", " ")
         if spec.camera == "top_down":
+            overhead_contract = top_down_subject_contract(spec)
             return compact_prompt_words(
-                f"Strict orthographic top-down game sprite. Camera at 90 degrees directly above the {subject}, "
-                "looking straight down. Show only top-facing surfaces and a compact foreshortened silhouette. "
-                "No visible frontal face, chest, facade, or side elevation. "
-                f"{description}. Exactly one complete isolated subject, centered on uniform white. "
-                f"{variation}Crisp game-ready details. No scene.",
+                f"Strict orthographic top-down game sprite of {description}. "
+                f"Camera exactly 90 degrees above, looking straight down. {overhead_contract} "
+                f"Exactly one complete isolated subject on uniform white. {variation}No scene.",
                 55,
             )
         return (