Add shape-aware sprite families

app.py

@@ -592,6 +592,100 @@ def draw_sprite(spec: AssetSpec, rng: random.Random) -> bytes:
     accent = colors[1] + (255,)
     trim = colors[3] + (255,)
 
+    if any(word in role_prompt for word in ("car", "racing", "racer", "buggy", "vehicle", "truck")):
+        shadow = (int(cx - 43 * scale), int(cy + 28 * scale), int(cx + 43 * scale), int(cy + 42 * scale))
+        draw.ellipse(shadow, fill=(0, 0, 0, 58))
+        draw.rounded_rectangle(
+            (cx - 34 * scale, cy - 42 * scale, cx + 34 * scale, cy + 42 * scale),
+            radius=int(18 * scale),
+            fill=outline,
+        )
+        draw.rounded_rectangle(
+            (cx - 26 * scale, cy - 34 * scale, cx + 26 * scale, cy + 34 * scale),
+            radius=int(13 * scale),
+            fill=body,
+        )
+        draw.rectangle((cx - 15 * scale, cy - 24 * scale, cx + 15 * scale, cy - 5 * scale), fill=accent)
+        draw.rectangle((cx - 18 * scale, cy + 8 * scale, cx + 18 * scale, cy + 27 * scale), fill=trim)
+        for side in (-1, 1):
+            draw.rounded_rectangle(
+                (cx + side * 27 * scale - 9 * scale, cy - 30 * scale, cx + side * 27 * scale + 9 * scale, cy - 13 * scale),
+                radius=int(5 * scale),
+                fill=(20, 22, 27, 255),
+            )
+            draw.rounded_rectangle(
+                (cx + side * 27 * scale - 9 * scale, cy + 15 * scale, cx + side * 27 * scale + 9 * scale, cy + 32 * scale),
+                radius=int(5 * scale),
+                fill=(20, 22, 27, 255),
+            )
+        out = io.BytesIO()
+        image.save(out, format="PNG")
+        return out.getvalue()
+
+    if any(word in role_prompt for word in ("turret", "tower", "cannon", "launcher")):
+        draw.ellipse((cx - 38 * scale, cy + 22 * scale, cx + 38 * scale, cy + 45 * scale), fill=(0, 0, 0, 60))
+        draw.ellipse((cx - 36 * scale, cy - 12 * scale, cx + 36 * scale, cy + 42 * scale), fill=outline)
+        draw.ellipse((cx - 28 * scale, cy - 5 * scale, cx + 28 * scale, cy + 34 * scale), fill=body)
+        barrel_angle = rng.choice((-0.45, -0.2, 0, 0.2, 0.45))
+        length = 52 * scale
+        bx = cx + math.sin(barrel_angle) * length
+        by = cy - 38 * scale
+        draw.line((cx, cy - 5 * scale, bx, by), fill=outline, width=max(10, int(17 * scale)))
+        draw.line((cx, cy - 5 * scale, bx, by), fill=trim, width=max(5, int(9 * scale)))
+        draw.rectangle((cx - 19 * scale, cy + 15 * scale, cx + 19 * scale, cy + 39 * scale), fill=accent)
+        out = io.BytesIO()
+        image.save(out, format="PNG")
+        return out.getvalue()
+
+    if role in ("coin", "gem", "key", "orb", "pickup", "collectible") or any(
+        word in role_prompt for word in ("coin", "gem", "key", "pickup", "collectible", "powerup", "power-up", "treasure")
+    ):
+        draw.ellipse((cx - 33 * scale, cy - 33 * scale, cx + 33 * scale, cy + 33 * scale), fill=outline)
+        if any(word in role_prompt for word in ("gem", "crystal", "diamond")):
+            points = [
+                (cx, cy - 43 * scale),
+                (cx + 36 * scale, cy - 8 * scale),
+                (cx + 20 * scale, cy + 40 * scale),
+                (cx - 20 * scale, cy + 40 * scale),
+                (cx - 36 * scale, cy - 8 * scale),
+            ]
+            draw.polygon(points, fill=body)
+            draw.polygon([(cx, cy - 34 * scale), (cx + 18 * scale, cy - 4 * scale), (cx, cy + 28 * scale), (cx - 18 * scale, cy - 4 * scale)], fill=accent)
+        elif "key" in role_prompt:
+            draw.ellipse((cx - 30 * scale, cy - 20 * scale, cx + 8 * scale, cy + 18 * scale), fill=body)
+            draw.ellipse((cx - 17 * scale, cy - 8 * scale, cx - 3 * scale, cy + 6 * scale), fill=(0, 0, 0, 0))
+            draw.rounded_rectangle((cx + 2 * scale, cy - 5 * scale, cx + 43 * scale, cy + 7 * scale), radius=int(5 * scale), fill=accent)
+            draw.rectangle((cx + 30 * scale, cy + 4 * scale, cx + 43 * scale, cy + 17 * scale), fill=trim)
+        else:
+            draw.ellipse((cx - 27 * scale, cy - 27 * scale, cx + 27 * scale, cy + 27 * scale), fill=body)
+            draw.ellipse((cx - 17 * scale, cy - 17 * scale, cx + 17 * scale, cy + 17 * scale), outline=trim, width=max(3, int(7 * scale)))
+            draw.arc((cx - 12 * scale, cy - 20 * scale, cx + 20 * scale, cy + 13 * scale), 195, 300, fill=(255, 255, 255, 180), width=max(2, int(4 * scale)))
+        out = io.BytesIO()
+        image.save(out, format="PNG")
+        return out.getvalue()
+
+    if any(word in role_prompt for word in ("slime", "blob", "gel", "jelly")):
+        draw.ellipse((cx - 40 * scale, cy + 24 * scale, cx + 40 * scale, cy + 42 * scale), fill=(0, 0, 0, 55))
+        draw.pieslice((cx - 40 * scale, cy - 28 * scale, cx + 40 * scale, cy + 52 * scale), 180, 360, fill=outline)
+        draw.pieslice((cx - 32 * scale, cy - 20 * scale, cx + 32 * scale, cy + 42 * scale), 180, 360, fill=body)
+        for side in (-1, 1):
+            draw.ellipse((cx + side * 11 * scale - 5 * scale, cy + 3 * scale, cx + side * 11 * scale + 5 * scale, cy + 13 * scale), fill=(255, 255, 255, 230))
+        draw.arc((cx - 12 * scale, cy + 12 * scale, cx + 12 * scale, cy + 27 * scale), 10, 170, fill=trim, width=max(2, int(4 * scale)))
+        out = io.BytesIO()
+        image.save(out, format="PNG")
+        return out.getvalue()
+
+    if any(word in role_prompt for word in ("beetle", "bug", "spider", "insect", "crawler", "creep")):
+        draw.ellipse((cx - 28 * scale, cy - 34 * scale, cx + 28 * scale, cy + 34 * scale), fill=outline)
+        draw.ellipse((cx - 21 * scale, cy - 27 * scale, cx + 21 * scale, cy + 29 * scale), fill=body)
+        draw.line((cx, cy - 25 * scale, cx, cy + 30 * scale), fill=trim, width=max(2, int(4 * scale)))
+        for y in (-18, 0, 18):
+            for side in (-1, 1):
+                draw.line((cx + side * 18 * scale, cy + y * scale, cx + side * 45 * scale, cy + (y + rng.choice((-10, 10))) * scale), fill=outline, width=max(3, int(5 * scale)))
+        out = io.BytesIO()
+        image.save(out, format="PNG")
+        return out.getvalue()
+
     if role in ("bullet", "projectile", "laser", "shot") or any(word in role_prompt for word in ("bullet", "projectile", "laser", "energy")):
         glow = colors[1] + (85,)
         core_width = rng.randint(7, 13) * scale
@@ -944,23 +1038,29 @@ def build_prompt_preview(specs: list[AssetSpec]) -> str:
     return "\n\n".join(f"{spec.role}:\n{spec.prompt}" for spec in specs)
 
 
+def build_model_report(rows: list[tuple[str, str]]) -> str:
+    return "\n".join(f"{role}: {source}" for role, source in rows)
+
+
 def generate_images_and_game(html_code: str, roles: str, style_hint: str):
     if not html_code.strip():
-        return "", "Paste HTML game code first.", [], "", ""
+        return "", "Paste HTML game code first.", [], "", "", ""
 
     specs = parse_assets(roles, style_hint or "pixel art style")
     if not specs:
-        return html_code, "Add at least one asset role, like `player: brave knight`.", [], "", build_preview(html_code)
+        return html_code, "Add at least one asset role, like `player: brave knight`.", [], "", "", build_preview(html_code)
 
     assets: dict[str, str] = {}
     gallery = []
     errors = []
+    model_rows = []
     run_id = time.time_ns()
 
     for index, spec in enumerate(specs):
         data_uri, gallery_path, error = generate_asset(spec, index, run_id)
         assets[spec.role] = data_uri
         gallery.append((gallery_path, f"{spec.role} -> {spec.filename}"))
+        model_rows.append((spec.role, HF_IMAGE_MODEL if error is None and HF_TOKEN else "local style fallback"))
         if error:
             errors.append(f"{spec.role}: fallback used ({error})")
 
@@ -970,7 +1070,7 @@ def generate_images_and_game(html_code: str, roles: str, style_hint: str):
     status = f"Generated and embedded {len(specs)} fresh asset(s) using {source}. Run {str(run_id)[-6:]}."
     if errors:
         status += "\n\n" + "\n".join(f"{item}: no HF image returned, used local style fallback" for item in [e.split(':', 1)[0] for e in errors])
-    return rewritten, status, gallery, build_prompt_preview(specs), build_preview(rewritten)
+    return rewritten, status, gallery, build_prompt_preview(specs), build_model_report(model_rows), build_preview(rewritten)
 
 
 with gr.Blocks(title="Image Generator for HTML Games") as demo:
@@ -1014,6 +1114,11 @@ with gr.Blocks(title="Image Generator for HTML Games") as demo:
                 lines=8,
                 interactive=False,
             )
+            model_report = gr.Textbox(
+                label="Model/source used by role",
+                lines=5,
+                interactive=False,
+            )
 
     gr.Markdown("## Game preview")
     preview = gr.HTML(build_preview(STARTER_HTML))
@@ -1021,7 +1126,7 @@ with gr.Blocks(title="Image Generator for HTML Games") as demo:
     generate_btn.click(
         fn=generate_images_and_game,
         inputs=[html_input, roles, style],
-        outputs=[output_code, status, gallery, prompt_preview, preview],
+        outputs=[output_code, status, gallery, prompt_preview, model_report, preview],
     )