Upload app.py

app.py

@@ -128,142 +128,145 @@ def build_event_sections(
     time_desc: str,
     season_desc: str,
     mood: str,
+    lat: float,
+    lon: float,
+    year: int,
 ) -> Dict[str, str]:
-    """Builds prompt sections focusing on event, location, year, and time."""
-    vocab = get_era_vocabulary(event.get("year") or 0)
-    region_ctx = get_region_context((event.get("facets") or {}).get("region"))
+    """Build VISUAL-FIRST event description for image generation.
+
+    Returns concrete visual elements that image models understand.
+    Target: 35-50 words for base prompt (style adds 15-20 more).
+    """
 
-    # Extract event details
     event_name = event.get("name", "Historical scene")
-    event_year = event.get("year", 0)
-    narrative = (event.get("narrative") or event.get("summary") or event.get("description") or "").strip()
+    event_year = event.get("year", year)
+    location = get_location_name(event, lat, lon)
+
+    # VISUAL SUBJECT (what the image shows)
+    subject_type = "historical scene"
+    if "battle" in event_name.lower() or "war" in event_name.lower():
+        subject_type = "battlefield"
+    elif "signing" in event_name.lower() or "declaration" in event_name.lower():
+        subject_type = "formal ceremony"
+    elif "speech" in event_name.lower() or "address" in event_name.lower():
+        subject_type = "public gathering"
+    elif "fall" in event_name.lower() or "liberation" in event_name.lower():
+        subject_type = "crowd scene"
+
+    # PARTICIPANTS (who's in the image with period clothing)
     actors = event.get("actors") or []
-    
-    # Get location context
-    region_name = (event.get("facets") or {}).get("region", "")
-    architecture = region_ctx.get("architecture") or vocab.get("architecture", "")
-    
-    # Build year/era marker (prominent)
-    if event_year < 0:
-        era_marker = f"{abs(event_year)} BCE"
-    elif event_year < 500:
-        era_marker = "ancient era"
-    elif event_year < 1500:
-        era_marker = "medieval era"
-    elif event_year < 1800:
-        era_marker = "early modern era"
-    elif event_year < 1900:
-        era_marker = "19th century"
-    else:
-        era_marker = f"{event_year}"
-    
-    # Get location name/region (prioritize location)
-    location_desc = ""
-    if region_name:
-        region_map = {
-            "western_europe": "Western Europe",
-            "eastern_europe": "Eastern Europe",
-            "north_america": "North America",
-            "south_america": "South America",
-            "east_asia": "East Asia",
-            "middle_east": "Middle East",
-            "africa": "Africa",
-        }
-        location_desc = region_map.get(region_name.lower(), region_name.replace("_", " ").title())
-    
-    # Simplify architecture description
-    architecture_lower = architecture.lower() if architecture else ""
-    if "roman" in architecture_lower:
-        arch_short = "Roman architecture"
-    elif "gothic" in architecture_lower:
-        arch_short = "Gothic architecture"
-    elif "medieval" in architecture_lower:
-        arch_short = "medieval architecture"
-    elif "asia" in region_name.lower() and event_year < 1900:
-        arch_short = "traditional Asian architecture"
-    else:
-        arch_short = architecture or "period architecture"
-    
-    # Extract key action from narrative (concise)
-    key_action = ""
-    if narrative:
-        # Get first short sentence or phrase
-        sentences = [s.strip() for s in narrative.split(".") if len(s.strip()) > 10 and len(s.strip()) < 80]
-        if sentences:
-            key_action = sentences[0]
-            # Trim if too long
-            if len(key_action) > 60:
-                key_action = key_action[:57] + "..."
-    
-    # Build focused scene: [Event] at [Location] in [Year] at [Time]: [Action]
-    # Make time prominent - include it in the main sentence
-    time_short = time_desc.split(",")[0] if "," in time_desc else time_desc  # Get first part of time description
-    
-    if location_desc:
-        if key_action:
-            scene_sentence = f"{event_name} at {location_desc} in {era_marker} at {time_short}: {key_action}"
-            else:
-            scene_sentence = f"{event_name} at {location_desc} in {era_marker} at {time_short}"
-        else:
-        if key_action:
-            scene_sentence = f"{event_name} in {era_marker} at {time_short}: {key_action}"
-else:
-            scene_sentence = f"{event_name} in {era_marker} at {time_short}"
-    
-    # Essential participants only (concise)
-    if actors and len(actors) > 0:
-        main_actors = actors[:2]  # Limit to 2 most important
-        if len(main_actors) == 1:
-            participants_sentence = f"{main_actors[0]} present"
-else:
-            participants_sentence = f"{main_actors[0]} and {main_actors[1]} present"
-        else:
-        participants_sentence = ""
-    
-    # Location architecture (concise, time already in main sentence)
-    if location_desc and arch_short:
-        location_sentence = f"{location_desc} with {arch_short}"
-    elif arch_short:
-        location_sentence = arch_short
+    participants_desc = ""
+    if actors:
+        # Add period-specific clothing descriptors
+        clothing = get_period_clothing(event_year)
+        if len(actors) == 1:
+            participants_desc = f"{actors[0]} in {clothing}"
+        elif len(actors) == 2:
+            participants_desc = f"{actors[0]} and {actors[1]} in {clothing}"
         else:
-        location_sentence = ""
-    
+            participants_desc = f"{actors[0]}, {actors[1]}, and others in {clothing}"
+
+    # ENVIRONMENT (where the scene takes place)
+    location_desc = get_environment_description(location, event_name, event_year)
+
+    # VISUAL ELEMENTS (period artifacts, architecture)
+    artifacts = event.get("artifacts") or []
+    visual_elements = join_list(artifacts[:3], "and") if artifacts else ""
+
+    # LIGHTING (time-based atmospheric description)
+    lighting = time_desc  # Use full description
+
     return {
-        "event_location_year": scene_sentence,  # Core: Event + Location + Year
-        "participants": participants_sentence,
-        "location_setting": location_sentence,
+        "subject": subject_type,
+        "participants": participants_desc,
+        "location": location_desc,
+        "elements": visual_elements,
+        "lighting": lighting,
+        "event_name": event_name,  # Keep for fallback
     }
 
 
+def get_period_clothing(year: int) -> str:
+    """Get period-appropriate clothing description."""
+    if year < 1500:
+        return "medieval robes and tunics"
+    elif year < 1700:
+        return "Renaissance doublets and robes"
+    elif year < 1800:
+        return "18th century coats and breeches"
+    elif year < 1850:
+        return "early 19th century military uniforms"
+    elif year < 1900:
+        return "Victorian formal attire"
+    elif year < 1920:
+        return "Edwardian formal dress"
+    elif year < 1950:
+        return "1940s military uniforms"
+    elif year < 1980:
+        return "mid-century formal wear"
+    else:
+        return "modern formal attire"
+
+
+def get_environment_description(location: str, event_name: str, year: int) -> str:
+    """Get concrete environmental description."""
+    if "°" in location:
+        env_name = ""
+    else:
+        env_name = location
+
+    if "battle" in event_name.lower() or "war" in event_name.lower():
+        if year < 1900:
+            return f"{env_name} countryside with period cannons, military encampments, and smoke from musket fire" if env_name else "muddy battlefield with period cannons, military encampments, and gunpowder smoke"
+        else:
+            return f"{env_name} terrain with military vehicles, fortifications, and artillery" if env_name else "war-torn terrain with military vehicles, fortifications, and artillery smoke"
+    elif "palace" in location.lower() or "hall" in event_name.lower():
+        return "grand ornate interior with period chandeliers, decorative architecture, and formal furnishings"
+    elif "street" in event_name.lower() or "crowd" in event_name.lower() or "fall" in event_name.lower():
+        return f"{env_name} streets with period buildings, gathered crowds, and urban architecture" if env_name else "city streets with period buildings, gathered crowds, and architectural details"
+    else:
+        return f"{env_name} with historically accurate period architecture and setting" if env_name else "period-accurate setting with appropriate historical architecture"
+
+
 def assemble_prompt_from_sections(sections: Dict[str, str], quality: str) -> str:
-    """Combines prompt sections into final prompt, keeping it 40-60 words."""
-    # Focused structure: [Event at Location in Year] [Participants] [Location/Setting] [Style/Quality]
-    parts = [
-        sections.get("event_location_year"),  # Core: Event + Location + Year (most important)
-        sections.get("participants"),  # Essential participants only
-        sections.get("location_setting"),  # Location architecture and time
-    ]
-    body = ". ".join(part for part in parts if part)
-    
-    # Add style/mood and quality tags at end (concise)
-    style_quality = f"{quality}"
-    result = f"{body}. {style_quality}".strip()
-    
-    # Ensure prompt is 40-60 words (target ~50 words)
-    word_count = len(result.split())
-    if word_count > 65:
-        # Trim if too long - keep core elements, reduce quality tags
-        words = result.split()
-        # Keep first 45 words (core content), add essential quality
-        core = " ".join(words[:45])
-        essential_quality = "historically accurate, photorealistic"
-        result = f"{core}. {essential_quality}"
-    elif word_count < 35:
-        # Expand slightly if too short - add more context
-        if not sections.get("participants"):
-            result = f"{body}. Historical figures visible. {style_quality}".strip()
-    
-    return result
+    """Assemble VISUAL-FIRST prompt for image generation.
+
+    Format: "[Subject] showing [participants]. [Environment with elements]. [Lighting]."
+    NO metadata - only visual descriptions that models understand.
+    """
+
+    subject = sections.get("subject", "historical scene")
+    participants = sections.get("participants", "")
+    location = sections.get("location", "")
+    elements = sections.get("elements", "")
+    lighting = sections.get("lighting", "")
+
+    # Build prompt sentence by sentence
+    parts = []
+
+    # Sentence 1: Subject + Participants
+    if participants:
+        parts.append(f"{subject.capitalize()} showing {participants}")
+    else:
+        event_name = sections.get("event_name", "historical event")
+        parts.append(f"{subject.capitalize()} depicting {event_name}")
+
+    # Sentence 2: Environment WITH Elements (combined)
+    if location and elements:
+        parts.append(f"{location} with {elements}")
+    elif location:
+        parts.append(location)
+    elif elements:
+        parts.append(f"Scene with {elements}")
+
+    # Sentence 3: Lighting/Atmosphere
+    if lighting:
+        parts.append(f"{lighting}")
+
+    # Join all parts with periods for clear structure
+    final = ". ".join(parts)
+
+    return final.strip()
 
 
 def build_fallback_prompt(
@@ -300,7 +303,7 @@ def build_fallback_prompt(
             "Glitch": "digital glitch art style, cyberpunk aesthetic, data corruption effects, historically themed"
         }
         quality = quality_map.get(mood, f"{mood} style, historically inspired")
-        else:
+    else:
         quality = f"{mood} style, historically accurate, photorealistic, 8K"
     prompt = assemble_prompt_from_sections(sections, quality)
     hint = (
@@ -344,14 +347,14 @@ print("🔑 API Token Status:")
 print("="*60)
 if HF_TOKEN:
     print(f"✅ HF_TOKEN: Found ({len(HF_TOKEN)} chars) - {HF_TOKEN[:10]}...")
-            else:
+else:
     print("❌ HF_TOKEN: Not found!")
     print("   Set HUGGINGFACE_API_TOKEN or HF_TOKEN environment variable")
 
 if GEMINI_API_KEY:
     print(f"✅ GEMINI_API_KEY: Found ({len(GEMINI_API_KEY)} chars) - {GEMINI_API_KEY[:10]}...")
     genai.configure(api_key=GEMINI_API_KEY)
-        else:
+else:
     print("⚠️  GEMINI_API_KEY: Not found (prompts will use fallback)")
 print("="*60 + "\n")
 
@@ -420,7 +423,7 @@ def generate_historical_prompt(
 
     if events:
         focus_event = events[0]
-        sections = build_event_sections(focus_event, time_desc, season_desc, mood)
+        sections = build_event_sections(focus_event, time_desc, season_desc, mood, lat, lon, year)
         # Style-appropriate quality tags
         if mood in ["Cartoon", "Minecraft", "Retro", "Glitch"]:
             quality_map = {
@@ -430,7 +433,7 @@ def generate_historical_prompt(
                 "Glitch": "digital glitch art style, cyberpunk aesthetic, data corruption effects, historically themed"
             }
             quality = quality_map.get(mood, f"{mood} style, historically inspired")
-            else:
+        else:
             quality = f"{mood} style, historically accurate, photorealistic, 8K"
         prompt = assemble_prompt_from_sections(sections, quality)
 
@@ -452,7 +455,7 @@ def generate_historical_prompt(
             year_match = f"📅 ~{year_delta} years"
         elif year_delta <= 10:
             year_match = f"📅 ~{year_delta} years apart"
-                            else:
+        else:
             year_match = f"⚠️ {year_delta} years apart"
 
         hint_lines = [
@@ -555,7 +558,7 @@ def process_coordinates(lat: float, lon: float, year: int, month: int, day: int,
         prompt = custom_prompt.strip()
         hint = f"📍 {lat:.4f}, {lon:.4f} | 📅 {year}-{month:02d}-{day:02d} {hour}:00 (Custom prompt)"
         status_parts.append("✅ Using custom prompt")
-        else:
+    else:
         status_parts.append("🔍 Searching historical events...")
         prompt, hint = generate_historical_prompt(lat, lon, year, month, day, hour, mood)
         status_parts.append("✅ Prompt generated")
@@ -592,9 +595,9 @@ def process_coordinates(lat: float, lon: float, year: int, month: int, day: int,
                 year_badge = "🎯"
             elif year_delta <= 5:
                 year_badge = "📅"
-        else:
+            else:
                 year_badge = "⏳"
-            
+
             timeline_md += (
                 f"**{event.get('year')}** {year_badge} — {source_icon} {event.get('name')}{qid_link} "
                 f"({event.get('distance_km')}km"
@@ -611,7 +614,7 @@ def process_coordinates(lat: float, lon: float, year: int, month: int, day: int,
             participants = event.get("actors") or event.get("participants") or []
             if participants and source == "wikidata":
                 timeline_md += f"_Participants: {', '.join(participants[:4])}_\n\n"
-            else:
+        else:
             timeline_md += "_No specific events found in database or Wikidata. Scene generated from era-appropriate context._"
     
     return image, prompt, hint, status, timeline_md
@@ -1065,9 +1068,9 @@ def create_app():
                 )
                 
                 gr.Markdown("### 🖼️ Generated Image")
-                        image_output = gr.Image(
-                    label="", 
-                            show_label=False,
+                image_output = gr.Image(
+                    label="",
+                    show_label=False,
                     height=600,
                     type="pil"
                 )
@@ -1147,7 +1150,7 @@ def create_app():
                         f"🧭 Parsed from prompt (confidence {parsed.confidence:.2f}) · "
                         f"{lat_val:.4f}, {lon_val:.4f}, year {year_val}"
                     )
-                    else:
+            else:
                     derived_note = "⚠️ Could not confidently parse prompt context; using manual inputs."
 
             image, prompt, hint, status, timeline = process_coordinates(