Update core/prompt_engineering.py

core/prompt_engineering.py

@@ -1,4 +1,10 @@
 # core/prompt_engineering.py
+import json
+
+# create_story_breakdown_prompt can remain mostly the same, but ensure 'key_action' is descriptive.
+# If key_action is very short, text overlays on video might not be useful.
+# Consider making 'key_action' slightly more verbose or adding a 'video_overlay_text' field.
+# For now, we'll use key_action.
 
 def create_story_breakdown_prompt(user_idea, genre="sci-fi", mood="suspenseful", num_scenes=3):
     return f"""
@@ -8,9 +14,9 @@ def create_story_breakdown_prompt(user_idea, genre="sci-fi", mood="suspenseful",
 
     Generate a {num_scenes}-scene story breakdown. For each scene, provide:
     1. scene_number (int)
-    2. setting_description (str): Vivid description of the location and atmosphere.
+    2. setting_description (str): Vivid description of the location and atmosphere (approx 20-40 words).
     3. characters_involved (list of str): Names of characters in the scene.
-    4. key_action (str): The main event or action happening.
+    4. key_action (str): The main event or action happening in one concise sentence (approx 10-20 words). This will be used for video overlays.
     5. dialogue_snippet (str): A brief, impactful line of dialogue if applicable.
     6. visual_style_suggestion (str): e.g., "Dark and gritty, high contrast, Blade Runner-esque neon"
     7. camera_angle_suggestion (str): e.g., "Low-angle shot to emphasize power"
@@ -20,13 +26,13 @@ def create_story_breakdown_prompt(user_idea, genre="sci-fi", mood="suspenseful",
     Example for one scene:
     {{
       "scene_number": 1,
-      "setting_description": "A dimly lit, cluttered spaceship cockpit. Warning lights flash intermittently.",
+      "setting_description": "A dimly lit, cluttered spaceship cockpit. Warning lights flash intermittently. Steam vents from a broken pipe.",
       "characters_involved": ["Captain Eva Rostova"],
-      "key_action": "Eva notices an unusual energy signature on the main console.",
-      "dialogue_snippet": "Eva: 'What in the void is that?'",
-      "visual_style_suggestion": "Claustrophobic, practical lighting, lens flares.",
-      "camera_angle_suggestion": "Close-up on Eva's face, then a point-of-view shot of the console.",
-      "emotional_beat": "Curiosity mixed with apprehension."
+      "key_action": "Eva frantically works at a console, trying to divert a catastrophic system failure.",
+      "dialogue_snippet": "Eva: 'Come on, come on... don't do this to me now!'",
+      "visual_style_suggestion": "Claustrophobic, practical lighting, lens flares, metallic sheens.",
+      "camera_angle_suggestion": "Close-up on Eva's determined face, sweat beading on her forehead.",
+      "emotional_beat": "Desperation and intense focus."
     }}
 
     Provide the full JSON structure for {num_scenes} scenes in a list:
@@ -37,83 +43,91 @@ def create_story_breakdown_prompt(user_idea, genre="sci-fi", mood="suspenseful",
     ]
     """
 
-def create_image_prompt_from_scene_data(scene_data, character_details=None, style_reference_desc=None):
+def create_image_prompt_from_scene_data(scene_data, character_definitions=None, style_reference_desc=None):
     """
-    Generates an image prompt from structured scene data.
+    Generates an image prompt from structured scene data, injecting character details.
     scene_data: dictionary for a single scene.
-    character_details: dict {char_name: description} for consistency (future use).
-    style_reference_desc: textual description of a desired style (future use).
+    character_definitions: dict {char_name_lower: description} for consistency.
+    style_reference_desc: textual description of a desired style.
     """
-    base_desc = f"Scene {scene_data.get('scene_number', '')}: {scene_data.get('key_action', '')}. Setting: {scene_data.get('setting_description', '')}."
+    setting_desc = scene_data.get('setting_description', '')
+    key_action_desc = scene_data.get('key_action', '')
     visual_style = scene_data.get('visual_style_suggestion', 'cinematic')
     camera_angle = scene_data.get('camera_angle_suggestion', '')
-
-    char_info = ""
-    if character_details and scene_data.get('characters_involved'):
-        for char_name in scene_data.get('characters_involved'):
-            if char_name in character_details:
-                char_info += f" {char_name} ({character_details[char_name]})."
-
+    emotional_beat = scene_data.get('emotional_beat', '')
+    
+    characters_str_parts = []
+    if character_definitions and scene_data.get('characters_involved'):
+        for char_name in scene_data.get('characters_involved', []):
+            # Match character name case-insensitively, but use original name from scene_data
+            defined_desc = character_definitions.get(char_name.lower().strip())
+            if defined_desc:
+                characters_str_parts.append(f"{char_name.strip()} ({defined_desc})")
+            else:
+                characters_str_parts.append(char_name.strip()) # Character mentioned but not defined
+    elif scene_data.get('characters_involved'): # Characters mentioned but no definitions provided
+        characters_str_parts = [name.strip() for name in scene_data.get('characters_involved', [])]
+
+    characters_involved_str = ""
+    if characters_str_parts:
+        if len(characters_str_parts) == 1:
+            characters_involved_str = f" The scene features {characters_str_parts[0]}."
+        else:
+            characters_involved_str = f" The scene features {', '.join(characters_str_parts[:-1])} and {characters_str_parts[-1]}."
+
+    base_desc = f"Depict: Scene {scene_data.get('scene_number', '')}. {key_action_desc} {characters_involved_str} Setting: {setting_desc}."
+    
     style_mod = ""
     if style_reference_desc:
         style_mod = f" Artistic style inspired by: {style_reference_desc}."
 
+    # Constructing a more robust prompt for DALL-E 3
+    # DALL-E 3 benefits from descriptive, story-like prompts.
     full_prompt = f"""
-    Generate a highly detailed, photorealistic image generation prompt.
-    The image should depict: {base_desc}{char_info}
-    Visual Style: {visual_style}{style_mod}.
+    Generate a highly detailed, photorealistic and cinematic image.
+    Image Description: {base_desc}
+    Visual Style: {visual_style}. {style_mod}
     Camera Perspective: {camera_angle}.
-    Emphasize mood: {scene_data.get('emotional_beat', '')}.
-    Focus on cinematic composition, lighting. Suitable for DALL-E 3 or Midjourney.
-    Output only the prompt string.
+    Emotional Tone: {emotional_beat}.
+    Key elements to emphasize: Cinematic composition, dramatic lighting, depth of field, rich textures, and atmospheric effects.
+    Output a visually stunning image suitable for a film storyboard.
     """
-    return full_prompt.strip()
-
+    # Removed "Output only the prompt string" as this function *is* the prompt string builder.
+    # DALL-E 3 does not need "Suitable for DALL-E 3 or Midjourney".
+    return " ".join(full_prompt.split()) # Cleans up extra whitespace
 
 def create_scene_regeneration_prompt(original_scene_data, user_feedback, full_story_context=None):
-    """
-    Creates a prompt for Gemini to regenerate a specific scene's script details.
-    original_scene_data: The JSON object of the scene to be modified.
-    user_feedback: Text from the user describing desired changes.
-    full_story_context: (Optional) JSON of all scenes for better consistency.
-    """
+    # ... (remains the same as your last working version) ...
     context_str = f"Original scene details:\n{json.dumps(original_scene_data, indent=2)}\n\n"
     if full_story_context:
         context_str += f"Full story context (this scene is number {original_scene_data.get('scene_number')}):\n{json.dumps(full_story_context, indent=2)}\n\n"
-
     return f"""
     You are an expert script doctor.
     {context_str}
     The user wants to modify this scene based on the following feedback: "{user_feedback}"
-
     Please regenerate ONLY the JSON object for this single scene, incorporating the feedback.
-    Maintain the same JSON structure as the original scene:
-    (scene_number, setting_description, characters_involved, key_action, dialogue_snippet, visual_style_suggestion, camera_angle_suggestion, emotional_beat).
-    Ensure the scene_number remains the same.
-    Focus on making the changes impactful while keeping the scene coherent.
-    If the feedback is about visual style or camera, update those fields accordingly.
-    If the feedback is about action or dialogue, update those.
+    Maintain the same JSON structure as the original scene (scene_number, setting_description, characters_involved, key_action, dialogue_snippet, visual_style_suggestion, camera_angle_suggestion, emotional_beat).
+    Ensure the scene_number remains the same. The 'key_action' should be a concise descriptive sentence (10-20 words).
     """
 
 def create_visual_regeneration_prompt(original_image_prompt, user_feedback_on_visuals, scene_data):
-    """
-    Creates a prompt for Gemini to regenerate an image prompt based on feedback.
-    """
+    # ... (remains the same, this prompt is for Gemini to *rewrite* an image prompt) ...
+    # This prompt should also be updated to be DALL-E 3 friendly if it's rewriting.
     return f"""
-    The previous image prompt for a scene was: "{original_image_prompt}"
+    The previous detailed image generation prompt for a scene was: "{original_image_prompt}"
     The scene details are:
     Setting: {scene_data.get('setting_description')}
     Action: {scene_data.get('key_action')}
+    Characters: {', '.join(scene_data.get('characters_involved',[]))}
     Mood/Emotion: {scene_data.get('emotional_beat')}
-    Visual Style: {scene_data.get('visual_style_suggestion')}
-    Camera: {scene_data.get('camera_angle_suggestion')}
+    Current Visual Style: {scene_data.get('visual_style_suggestion')}
+    Current Camera: {scene_data.get('camera_angle_suggestion')}
 
     The user provided this feedback on the visuals: "{user_feedback_on_visuals}"
 
-    Generate a new, refined image generation prompt that incorporates this feedback.
-    The new prompt should aim to correct or enhance the visuals as per the user's request.
+    Generate a new, refined, highly detailed, photorealistic and cinematic image generation prompt based on this feedback.
+    The new prompt should aim to correct or enhance the visuals as per the user's request, while maintaining the core scene elements.
+    Ensure the prompt is descriptive and suitable for generating a stunning image for a film storyboard with DALL-E 3.
+    Focus on cinematic composition, dramatic lighting, depth of field, rich textures, and atmospheric effects.
     Output only the new prompt string.
-    """
-
-# (Keep other prompt functions if needed)
-import json # Add this at the top of prompt_engineering.py if not already there
+    """