Update utils.py

utils.py

@@ -150,10 +150,10 @@ def apply_flux_rules(prompt: str, analysis_metadata: Optional[Dict[str, Any]] =
     # Extract description part only (remove CAMERA_SETUP section if present)
     description_part = _extract_description_only(cleaned_prompt)
     
-    # NEW: Condense the description to remove redundancy
+    # NEW: Convert from descriptive to generative language
     if PROFESSIONAL_PHOTOGRAPHY_CONFIG.get("prompt_condensation", True):
-        description_part = _condense_description(description_part)
-        logger.info("Applied prompt condensation")
+        description_part = _convert_to_generative_language(description_part)
+        logger.info("Converted to generative language")
     
     # Check if BAGEL provided intelligent camera setup with cinematography context
     camera_config = ""
@@ -224,44 +224,89 @@ def _extract_description_only(prompt: str) -> str:
     return description.strip()
 
 
-def _condense_description(description: str) -> str:
-    """Condense BAGEL's verbose description into concise prompt format"""
+def _convert_to_generative_language(description: str) -> str:
+    """Convert descriptive analysis language to direct generative prompt language"""
     try:
-        condensed = description
-        
-        # Remove redundant phrases
-        redundant_phrases = FLUX_RULES.get("condensation_patterns", {}).get("remove_redundant_phrases", [])
-        for phrase in redundant_phrases:
-            condensed = re.sub(phrase, '', condensed, flags=re.IGNORECASE)
-        
-        # Simplify lighting descriptions
-        lighting_simplifications = FLUX_RULES.get("condensation_patterns", {}).get("simplify_lighting_descriptions", [])
-        for pattern, replacement in lighting_simplifications:
-            condensed = re.sub(pattern, replacement, condensed, flags=re.IGNORECASE)
-        
-        # Condense mood descriptions
-        mood_condensations = FLUX_RULES.get("condensation_patterns", {}).get("condense_mood_descriptions", [])
-        for pattern, replacement in mood_condensations:
-            condensed = re.sub(pattern, replacement, condensed, flags=re.IGNORECASE)
-        
-        # Remove excessive adjectives and verbose explanations
-        condensed = re.sub(r'\s*,\s*which\s+[^,]+,\s*', ', ', condensed)
-        condensed = re.sub(r'\s*\.\s*The\s+', '. ', condensed)
-        condensed = re.sub(r'\s+', ' ', condensed)
-        
-        # Split into sentences and keep only the most essential
-        sentences = [s.strip() for s in condensed.split('.') if s.strip()]
-        
-        # Keep first 2-3 most descriptive sentences
-        if len(sentences) > 3:
-            essential_sentences = sentences[:3]
-            condensed = '. '.join(essential_sentences)
-        
-        logger.info(f"Condensed description from {len(description)} to {len(condensed)} characters")
-        return condensed.strip()
+        generative = description
+        
+        # Remove descriptive introduction phrases
+        descriptive_intros = [
+            r'This image (?:features|shows|depicts|presents|displays)',
+            r'The image (?:features|shows|depicts|presents|displays)',
+            r'This (?:photograph|picture|scene|composition) (?:features|shows|depicts)',
+            r'The (?:photograph|picture|scene|composition) (?:features|shows|depicts)',
+            r'This is (?:a|an) (?:image|photograph|picture) (?:of|showing)',
+            r'The setting (?:appears to be|is)',
+            r'The scene (?:appears to be|is|shows)',
+            r'(?:In the background|In the foreground), (?:there are|there is)',
+            r'(?:The background|The foreground) (?:features|shows|contains)',
+            r'(?:There are|There is) [^,]+ (?:in the background|in the foreground)',
+            r'The overall (?:setting|atmosphere|mood) (?:suggests|indicates)',
+        ]
+        
+        for pattern in descriptive_intros:
+            generative = re.sub(pattern, '', generative, flags=re.IGNORECASE)
+        
+        # Remove uncertainty phrases
+        uncertainty_phrases = [
+            r'possibly (?:a|an) ',
+            r'appears to be (?:a|an) ',
+            r'seems to be (?:a|an) ',
+            r'might be (?:a|an) ',
+            r'could be (?:a|an) ',
+            r'suggests (?:a|an) ',
+            r'indicating (?:a|an) ',
+            r'(?:possibly|apparently|seemingly|likely)',
+        ]
+        
+        for pattern in uncertainty_phrases:
+            generative = re.sub(pattern, '', generative, flags=re.IGNORECASE)
+        
+        # Convert descriptive structure to noun phrases
+        structural_conversions = [
+            # "close-up of a X" -> "close-up X"
+            (r'(?:close-up|medium shot|wide shot) of (?:a|an|the) ', r'close-up '),
+            # "blurred figures of people" -> "blurred people"
+            (r'(?:blurred )?(?:figures|silhouettes) of (\w+)', r'blurred \1'),
+            # "people walking on a sidewalk" -> "people walking on sidewalk"
+            (r'(?:a|an|the) (\w+)', r'\1'),
+            # Remove excessive connecting words
+            (r'(?:, and|, with|, featuring)', ','),
+            # Simplify location descriptions
+            (r'on (?:a|an|the) ', r'on '),
+            (r'in (?:a|an|the) ', r'in '),
+        ]
+        
+        for pattern, replacement in structural_conversions:
+            generative = re.sub(pattern, replacement, generative, flags=re.IGNORECASE)
+        
+        # Convert action descriptions to present participles
+        action_conversions = [
+            (r'(\w+) (?:are|is) walking', r'\1 walking'),
+            (r'(\w+) (?:are|is) standing', r'\1 standing'),
+            (r'(\w+) (?:are|is) sitting', r'\1 sitting'),
+            (r'people (?:are|is) out of focus', r'blurred people'),
+        ]
+        
+        for pattern, replacement in action_conversions:
+            generative = re.sub(pattern, replacement, generative, flags=re.IGNORECASE)
+        
+        # Clean up extra spaces and punctuation
+        generative = re.sub(r'\s+', ' ', generative)
+        generative = re.sub(r'^\s*,\s*', '', generative)  # Remove leading commas
+        generative = re.sub(r'\s*,\s*,+', ',', generative)  # Remove double commas
+        generative = re.sub(r'\.+', '.', generative)  # Remove multiple periods
+        
+        # Ensure it starts with a capital letter
+        generative = generative.strip()
+        if generative:
+            generative = generative[0].upper() + generative[1:] if len(generative) > 1 else generative.upper()
+        
+        logger.info(f"Converted descriptive to generative: {len(description)} → {len(generative)} chars")
+        return generative
         
     except Exception as e:
-        logger.warning(f"Description condensation failed: {e}")
+        logger.warning(f"Generative language conversion failed: {e}")
         return description
 
 
@@ -334,6 +379,8 @@ def _optimize_prompt_length(prompt: str) -> str:
     except Exception as e:
         logger.warning(f"Prompt length optimization failed: {e}")
         return prompt
+
+
 def _detect_scene_from_description(description_lower: str) -> str:
     """Enhanced scene detection from description with cinematography knowledge"""
     scene_keywords = PROFESSIONAL_PHOTOGRAPHY_CONFIG.get("scene_detection_keywords", {})
@@ -437,12 +484,12 @@ def _get_enhanced_camera_config(scene_type: str, description_lower: str) -> str:
     """Get enhanced camera configuration with cinematography knowledge"""
     # Enhanced camera configurations with cinema equipment
     enhanced_configs = {
-        "cinematic": ", Shot on ARRI Alexa LF, 35mm anamorphic lens, cinematic photography",
-        "portrait": ", Shot on Canon EOS R5, 85mm f/1.4 lens at f/2.8, professional portrait photography",
-        "landscape": ", Shot on Phase One XT, 24-70mm f/4 lens at f/8, epic landscape photography",
-        "street": ", Shot on Leica M11, 35mm f/1.4 lens at f/2.8, documentary street photography",
-        "architectural": ", Shot on Canon EOS R5, 24-70mm f/2.8 lens at f/8, architectural photography",
-        "commercial": ", Shot on Hasselblad X2D 100C, 90mm f/2.5 lens, commercial photography"
+        "cinematic": ", Shot on ARRI Alexa LF, 35mm anamorphic lens at f/2.8, ISO 400, cinematic framing, film grain, dramatic composition",
+        "portrait": ", Shot on Canon EOS R5, 85mm f/1.4 lens at f/2.8, ISO 200, rule of thirds composition, professional portrait photography",
+        "landscape": ", Shot on Phase One XT, 24-70mm f/4 lens at f/8, ISO 100, hyperfocal distance, leading lines composition, epic landscape photography",
+        "street": ", Shot on Leica M11, 35mm f/1.4 lens at f/2.8, ISO 800, decisive moment, candid composition, documentary street photography",
+        "architectural": ", Shot on Canon EOS R5, 24-70mm f/2.8 lens at f/8, ISO 100, symmetrical composition, perspective correction, architectural photography",
+        "commercial": ", Shot on Hasselblad X2D 100C, 90mm f/2.5 lens at f/4, ISO 100, centered composition, product photography"
     }
     
     # Use enhanced config if available, otherwise fall back to FLUX_RULES
@@ -482,7 +529,7 @@ def _get_style_enhancement(scene_type: str, description_lower: str) -> str:
         if "film grain" not in description_lower:
             return ", " + style_enhancements.get("cinematic", "cinematic composition, film grain")
     elif scene_type in ["portrait", "commercial"]:
-        return ", " + style_enhancements.get("photorealistic", "photorealistic, ultra-detailed")
+        return ", " + style_enhancements.get("photorealistic", "photorealistic rendering, ultra-detailed")
     elif "editorial" in description_lower:
         return ", " + style_enhancements.get("editorial", "editorial photography style")