Enforce top-down sprite perspective

app.py

@@ -50,6 +50,7 @@ class AssetSpec:
     variation: str = ""
     expected_subjects: int | None = 1
     silhouette: str = "any"
+    description: str = ""
 
 
 @dataclass
@@ -183,6 +184,23 @@ def compact_prompt_words(text: str, limit: int) -> str:
     return " ".join(words[:limit])
 
 
+def compact_subject_description(text: str, limit: int) -> str:
+    """Keep user subject details while removing camera/composition words supplied elsewhere."""
+    value = text or ""
+    value = re.sub(r"^\s*(?:one\s+single|single|one)\s+", "", value, flags=re.I)
+    value = re.sub(
+        r"\b(?:viewed|seen)\s+(?:directly\s+)?from\s+(?:an?\s+)?(?:overhead|above)\b",
+        " ",
+        value,
+        flags=re.I,
+    )
+    value = re.sub(r"\b(?:strict\s+)?(?:top[- ]down|overhead)\b", " ", value, flags=re.I)
+    value = re.sub(r"\b(?:fully\s+)?isolated\b", " ", value, flags=re.I)
+    value = re.sub(r"\s+([,.;:])", r"\1", value)
+    value = re.sub(r"([,.;:])(?:\s*[,.;:])+", r"\1", value)
+    return compact_prompt_words(value, limit)
+
+
 def primary_camera_phrase(camera: str, is_background: bool) -> str:
     if camera == "top_down":
         return "Strict orthographic overhead view directly from above."
@@ -718,6 +736,7 @@ def parse_assets(
                     filename=filename,
                     width=width,
                     height=height,
+                    description=interpreted_prompt,
                     camera=camera,
                     group=group,
                     variant_index=variant_index,
@@ -1758,18 +1777,21 @@ def diffusion_dimensions(spec: AssetSpec) -> tuple[int, int]:
 
 def diffusion_negative_prompt(spec: AssetSpec) -> str:
     camera_negative = {
-        "top_down": ", sky, horizon, eye-level view, front view, front-facing facade, vanishing point",
+        "top_down": (
+            ", sky, horizon, eye-level view, front view, front-facing facade, upright portrait, "
+            "standing toward camera, face looking at viewer, vanishing point"
+        ),
         "isometric": ", eye-level view, front view, inconsistent perspective, horizon",
         "side_view": ", overhead view, top-down view, isometric view",
     }.get(spec.camera, "")
     if spec.composition in {"single_subject", "icon", "animation_frame"}:
-        return (
-            "multiple subjects, duplicate character, repeated subject, character sheet, turnaround, lineup, alternate views, "
-            "multiple poses, cropped subject, scenery, landscape, environment, environmental framing, ground plane, "
-            "background props, drop shadow, text, watermark, blurry, soft focus, out of focus, low detail, smeared details, "
-            "compression artifacts"
-            + camera_negative
+        negative = (
+            "multiple subjects, duplicate character, repeated subject, character sheet, turnaround, lineup, "
+            "alternate views, multiple poses, cropped subject, scenery, landscape, environment, ground plane, "
+            "background props, drop shadow, text, watermark, blurry, soft focus, out of focus, low detail"
         )
+        camera_prefix = camera_negative.lstrip(", ")
+        return compact_prompt_words(f"{camera_prefix}, {negative}" if camera_prefix else negative, 55)
     if spec.composition == "seamless":
         return "visible seams, borders, frame, perspective mockup, text, watermark, blurry, soft focus, low detail" + camera_negative
     if spec.composition == "sprite_sheet":
@@ -1920,9 +1942,17 @@ def primary_diffusion_prompt(spec: AssetSpec) -> str:
         camera = f"User-defined camera: {compact_prompt_words(spec.camera_instruction, 10)}."
     variation = f"Distinct variation: {compact_prompt_words(spec.variation, 8)}. " if spec.variation else ""
     if spec.composition == "single_subject":
-        description = compact_prompt_words(spec.prompt, 6 if variation else 10)
+        description = compact_subject_description(spec.description or spec.prompt, 10 if variation else 16)
         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":
+            return compact_prompt_words(
+                f"Top-down game sprite. Camera directly above the {subject}, looking straight down. "
+                "Show top surfaces and a compact foreshortened silhouette, never an upright front-facing portrait. "
+                f"{description}. Exactly one complete isolated subject, centered on uniform white. "
+                f"{variation}Crisp game-ready details. No scene.",
+                55,
+            )
         return (
             f"{camera} Single isolated {subject} subject. One complete centered body in one pose. "
             "Fill eighty percent of the frame with even margins. "
@@ -1944,7 +1974,13 @@ def primary_diffusion_prompt(spec: AssetSpec) -> str:
 
 def primary_generation_dimensions(spec: AssetSpec) -> tuple[int, int]:
     if not is_background_spec(spec):
-        return 768, 1024
+        if spec.camera in {"top_down", "isometric"}:
+            return 1024, 1024
+        if spec.camera == "side_view":
+            return 1152, 896
+        if spec.camera in {"front_view", "first_person"}:
+            return 896, 1152
+        return 1024, 1024
     aspect = spec.width / max(1, spec.height)
     if aspect >= 1.35:
         return 1344, 768
@@ -2016,6 +2052,11 @@ def primary_diffusion_png(spec: AssetSpec, index: int, run_id: int) -> tuple[byt
             if has_implausibly_thin_foreground_subject(content, spec):
                 last_failure_detail = "the last output contained an implausibly thin foreground silhouette"
                 continue
+            if has_front_view_like_top_down_silhouette(content, spec):
+                last_failure_detail = (
+                    "the last output retained a tall portrait-like silhouette instead of the requested overhead view"
+                )
+                continue
             if has_undersized_foreground_subject(content):
                 last_failure_detail = "the last output foreground occupied too little of the sprite canvas"
                 continue
@@ -2359,6 +2400,65 @@ def has_implausibly_thin_foreground_subject(content: bytes, spec: AssetSpec) ->
     )
 
 
+def has_front_view_like_top_down_silhouette(content: bytes, spec: AssetSpec) -> bool:
+    """Conservatively reject upright portrait silhouettes for overhead sprite contracts.
+
+    Alpha geometry cannot prove camera perspective, but a tall, canvas-filling
+    humanoid is a strong signal that the generator ignored an overhead request.
+    Broader non-humanoid checks apply only when the user's own description asks
+    for a radial, wide, or multi-legged body. Narrow top-down props and vehicles
+    therefore remain valid user-configurable assets.
+    """
+    if spec.camera != "top_down" or spec.composition not in {"single_subject", "icon", "animation_frame"}:
+        return False
+    image = Image.open(io.BytesIO(content)).convert("RGBA")
+    bbox = material_foreground_bbox_from_image(image)
+    if bbox is None:
+        return False
+    x0, y0, x1, y1 = bbox
+    bbox_width = x1 - x0 + 1
+    bbox_height = y1 - y0 + 1
+    if bbox_width < 1:
+        return True
+    aspect = bbox_height / bbox_width
+    height_ratio = bbox_height / max(1, image.height)
+    if height_ratio < 0.72:
+        return False
+
+    if spec.silhouette == "humanoid" and aspect >= 1.35:
+        return True
+
+    description = f"{spec.description} {spec.prompt}"
+    broad_shape_requested = contains_any_term(
+        description,
+        (
+            "broad",
+            "radial",
+            "wide-bodied",
+            "wide bodied",
+            "six-legged",
+            "six legged",
+            "four-legged",
+            "four legged",
+            "multi-legged",
+            "multi legged",
+            "quadruped",
+            "spider-like",
+            "spider like",
+            "crab-like",
+            "crab like",
+        ),
+    )
+    if broad_shape_requested and aspect >= 1.45:
+        return True
+
+    narrow_subject_requested = contains_any_term(
+        description,
+        ("missile", "rocket", "sword", "spear", "beam", "projectile", "tower", "pole"),
+    )
+    return not narrow_subject_requested and aspect >= 2.0
+
+
 def has_undersized_foreground_subject(content: bytes) -> bool:
     """Detect sprites whose visible model foreground occupies too little of its game canvas."""
     image = Image.open(io.BytesIO(content)).convert("RGBA")
@@ -2673,6 +2773,8 @@ def validate_asset_png(content: bytes, spec: AssetSpec) -> list[str]:
                 )
             elif spec.expected_subjects == 1 and has_implausibly_thin_foreground_subject(content, spec):
                 warnings.append("sprite foreground silhouette is implausibly thin")
+            elif spec.expected_subjects == 1 and has_front_view_like_top_down_silhouette(content, spec):
+                warnings.append("top-down sprite silhouette remains strongly portrait-like")
             elif spec.expected_subjects == 1 and has_undersized_foreground_subject(content):
                 warnings.append("foreground subject occupies too little of the output canvas")
     if spec.composition == "seamless" and image.width > 1 and image.height > 1: