Update core/prompt_engineering.py

core/prompt_engineering.py

@@ -1,114 +1,189 @@
 # core/prompt_engineering.py
 import json
 
-# create_story_breakdown_prompt - (ensure key_action and emotional_beat are descriptive) - REMAINS SAME
-def create_story_breakdown_prompt(user_idea, genre="sci-fi", mood="suspenseful", num_scenes=3):
+def create_cinematic_treatment_prompt(user_idea, genre, mood, num_scenes=3, creative_guidance="standard"):
+    """
+    Generates a prompt for Gemini to create a full cinematic treatment, including
+    proactive suggestions for visual style, camera, sound, and even thematic elements.
+    creative_guidance: "standard", "more_artistic", "experimental_narrative"
+    """
+    guidance_detail = {
+        "standard": "Provide solid, genre-appropriate suggestions.",
+        "more_artistic": "Lean into more artistic, unconventional, and visually striking suggestions for style and camera. Suggest unique color palettes or lighting.",
+        "experimental_narrative": "Feel free to suggest a minor unexpected narrative twist or a symbolic visual motif that could enhance the story within one of the scenes."
+    }[creative_guidance]
+
     return f"""
-    You are an expert screenwriter and visual storyteller. Based on: "{user_idea}", genre: "{genre}", mood: "{mood}".
-    Generate a {num_scenes}-scene story breakdown. For each scene:
-    1. scene_number (int)
-    2. emotional_beat (str): Short title for the scene's core feeling.
-    3. setting_description (str): Vivid description (30-50 words).
-    4. characters_involved (list of str): Names of characters.
-    5. key_action (str): Main event for video overlay (15-20 words).
-    6. dialogue_snippet (str): Brief impactful dialogue.
-    7. visual_style_suggestion (str): Keywords for visual style.
-    8. camera_angle_suggestion (str): Specific camera shot.
-    Output ONLY the JSON list of scenes. Example:
-    {{ "scene_number": 1, "emotional_beat": "Tense Standoff", "setting_description": "Rain-slicked, neon-drenched alleyway...", "characters_involved": ["Detective Kaito", "Informant"], "key_action": "Kaito cautiously approaches a nervous informant.", "dialogue_snippet": "Informant: 'They know...'", "visual_style_suggestion": "Neo-noir, cyberpunk...", "camera_angle_suggestion": "Medium shot..."}}
-    [{{"scene1_details..."}}, {{"scene2_details..."}}]
+    You are an AI Creative Director and Master Storyteller, collaborating on a cinematic concept.
+    Base Idea: "{user_idea}"
+    Genre: "{genre}"
+    Mood: "{mood}"
+    Number of Key Scenes: {num_scenes}
+    Creative Guidance Level: {creative_guidance} ({guidance_detail})
+
+    Task: Develop a rich cinematic treatment. For EACH of the {num_scenes} key scenes, provide:
+    1.  `scene_number` (int): Sequential.
+    2.  `scene_title` (str): A short, evocative title for the scene (e.g., "The Neon Rains of Sector 7", "Echoes in the Void").
+    3.  `setting_description` (str): Vivid, sensory details (sight, sound, atmosphere). Where are we? What makes it unique? (40-60 words).
+    4.  `characters_involved` (list of str): Names of characters central to this scene.
+    5.  `character_focus_moment` (str): For the primary character(s) in this scene, describe a key internal thought, subtle expression, or micro-action that reveals their state of mind or advances their arc.
+    6.  `key_plot_beat` (str): The most critical plot development or character action in this scene (1-2 sentences).
+    7.  `suggested_dialogue_hook` (str): One potent line of dialogue that captures the scene's essence or a character's voice. (If no dialogue, state "Silent scene").
+    8.  `PROACTIVE_visual_style_감독` (str): Your proactive, detailed suggestion for this scene's visual style. Go beyond generic terms. Think specific art movements, film references, color theory, lighting techniques (e.g., "Dutch angles with chiaroscuro lighting, using a desaturated palette with piercing cyan highlights, reminiscent of early Tarkovsky but with a cyberpunk edge").
+    9.  `PROACTIVE_camera_work_감독` (str): Your proactive suggestion for impactful camera work. Describe a specific shot or short sequence (e.g., "Slow dolly zoom into the protagonist's eyes, followed by a whip pan to reveal the approaching threat off-screen").
+    10. `PROACTIVE_sound_design_감독` (str): Key ambient sounds, specific SFX, and a suggestion for the musical mood/instrumentation for this scene (e.g., "Ambient: Distant city hum, dripping water. SFX: Glitching electronic spark. Music: Low, ominous synth pads with a recurring, detuned piano motif").
+    11. `dalle_image_prompt_keywords` (str): A concise list of 5-7 powerful keywords extracted from the above, specifically for generating a DALL-E image that captures the visual essence of this scene. Focus on nouns, strong adjectives, and artistic styles. (e.g., "cyberpunk alleyway, neon rain, lone figure, glowing data streams, high contrast, cinematic").
+    12. `pexels_search_query_감독` (str): A concise, effective search query (2-4 words) for Pexels to find a background or atmospheric shot relevant to this scene's setting or mood (e.g., "rainy neon city," "vast desert landscape," "server room interior").
+
+    If `creative_guidance` is "experimental_narrative", for one scene, you may subtly alter `key_plot_beat` or add a symbolic element to `setting_description` to introduce an unexpected twist, explaining your reasoning briefly in a `director_note` field for that scene only.
+
+    Output ONLY a valid JSON list of these scene objects. Ensure all field names are exactly as specified (감독 denotes your proactive directorial input).
     """
 
+# create_image_prompt_from_scene_data - THIS WILL NOW BE SIMPLER.
+# Gemini already gave us dalle_image_prompt_keywords. We can build upon that.
+def construct_dalle_prompt(scene_data, character_definitions=None, global_style_additions=""):
+    """
+    Constructs the final DALL-E prompt using keywords from Gemini,
+    character details, and global style.
+    """
+    base_keywords = scene_data.get('dalle_image_prompt_keywords', 'cinematic scene')
+    setting_desc = scene_data.get('setting_description', '') # For context
+    action_desc = scene_data.get('key_plot_beat', '') # For context
+    director_visual_style = scene_data.get('PROACTIVE_visual_style_감독', '')
+    director_camera = scene_data.get('PROACTIVE_camera_work_감독', '')
 
-# create_image_prompt_from_scene_data - (injects char defs & style) - REMAINS SAME
-def create_image_prompt_from_scene_data(scene_data, character_definitions=None, global_style_reference=""):
-    emotional_beat_title = scene_data.get('emotional_beat', 'A cinematic scene')
-    setting_desc = scene_data.get('setting_description', 'A visually interesting setting.')
-    key_action_desc = scene_data.get('key_action', 'A significant moment unfolds.')
-    characters_involved_in_scene = scene_data.get('characters_involved', [])
-    character_prompt_segments = []
-    if characters_involved_in_scene:
-        for char_name_from_scene in characters_involved_in_scene:
-            char_name_clean = char_name_from_scene.strip(); char_lookup_key = char_name_clean.lower()
+    character_details_for_prompt = []
+    # ... (Character injection logic remains the same as your last version, using char_name_lookup etc.)
+    if scene_data.get('characters_involved'):
+        for char_name_in_scene in scene_data.get('characters_involved', []):
+            char_name_clean = char_name_in_scene.strip(); char_lookup_key = char_name_clean.lower()
             if character_definitions and char_lookup_key in character_definitions:
-                character_prompt_segments.append(f"{char_name_clean} (described as: {character_definitions[char_lookup_key]})")
+                character_details_for_prompt.append(f"{char_name_clean} ({character_definitions[char_lookup_key]})")
             else: character_prompt_segments.append(char_name_clean)
-    characters_narrative = ""
-    if character_prompt_segments:
-        if len(character_prompt_segments) == 1: characters_narrative = f"The main character is {character_prompt_segments[0]}."
-        else: characters_narrative = f"The scene features {', '.join(character_prompt_segments[:-1])}, and {character_prompt_segments[-1]}."
-    narrative_prompt = f"Scene Number: {scene_data.get('scene_number', 'N/A')}. Setting: {setting_desc}. {characters_narrative} Key Action: {key_action_desc}. Emotional Tone: {scene_data.get('emotional_beat', '')}."
-    style_instructions = f"Visual Style: {scene_data.get('visual_style_suggestion', 'cinematic, photorealistic')}."
-    if global_style_reference: style_instructions += f" Specific style reference: {global_style_reference}."
-    camera_instructions = f"Camera Perspective: {scene_data.get('camera_angle_suggestion', 'eye-level medium shot')}."
-    full_prompt = (f"Generate an ultra-detailed, photorealistic, and highly cinematic digital painting or concept art image. "
-                   f"The image should depict: '{emotional_beat_title}'. Narrative Context: {narrative_prompt} "
-                   f"Artistic & Technical Instructions: {style_instructions} {camera_instructions} "
-                   f"Emphasize: Cinematic composition, dramatic lighting, rich textures, depth of field, strong atmospheric effects. "
-                   f"The image must feel like a high-quality film still. Pay close attention to character details.")
-    return " ".join(full_prompt.split())
-
-
-# --- NEW: Prompt for Narration Script ---
-def create_narration_script_prompt(story_scenes_data, overall_mood, overall_genre):
+    
+    character_narrative = ""
+    if character_details_for_prompt:
+        character_narrative = f" Characters featured: {', '.join(character_details_for_prompt)}."
+
+    # DALL-E 3 often prefers more natural language but benefits from strong keywords.
+    # We combine Gemini's suggestions with a structured approach.
+    prompt = (
+        f"Create an ultra-detailed, photorealistic, and highly cinematic masterpiece image. "
+        f"Theme: '{scene_data.get('scene_title', 'A dramatic moment')}'. "
+        f"Core visual elements based on keywords: {base_keywords}. " # Gemini's keywords
+        f"{character_narrative} " # Injected character descriptions
+        f"The scene unfolds in this setting: {setting_desc}. The key moment is: {action_desc}. "
+        f"Artistic Direction -- Visual Style: {director_visual_style}. {global_style_additions}. "
+        f"Cinematography -- Camera Work: {director_camera}. "
+        f"Overall Impression: Evoke the mood of '{scene_data.get('mood', 'intense')}' with an atmosphere of '{scene_data.get('emotional_beat', 'suspense')}'. "
+        f"Render with extreme detail, complex lighting, and rich textures suitable for a blockbuster film's concept art. "
+    )
+    return " ".join(prompt.split())
+
+
+def create_narration_script_prompt_enhanced(story_scenes_data, overall_mood, overall_genre, voice_style="cinematic_trailer"):
     """
-    Generates a prompt for Gemini to write a concise narration script for an animatic,
-    covering all provided scenes.
+    Generates a more nuanced narration script prompt for Gemini.
+    voice_style: "cinematic_trailer", "documentary_neutral", "introspective_character"
     """
+    # ... (scene summary logic remains the same as your last version) ...
     scenes_summary = []
     for i, scene in enumerate(story_scenes_data):
         scenes_summary.append(
-            f"Scene {scene.get('scene_number', i+1)} ({scene.get('emotional_beat','')}):\n"
-            f"- Setting: {scene.get('setting_description','')}\n"
-            f"- Key Action: {scene.get('key_action','')}\n"
-            f"- Characters: {', '.join(scene.get('characters_involved',[]))}\n"
-            f"- Implied Dialogue/Thought: {scene.get('dialogue_snippet','(none)')}"
+            f"Scene {scene.get('scene_number', i+1)} ({scene.get('scene_title','Untitled')} - {scene.get('emotional_beat','')})"
+            f": {scene.get('key_plot_beat','')} Focus: {scene.get('character_focus_moment','')} "
+            f"Soundscape hint: {scene.get('PROACTIVE_sound_design_감독','')}"
         )
-    
-    full_summary_text = "\n\n".join(scenes_summary)
+    full_summary_text = "\n".join(scenes_summary)
 
-    prompt = f"""
-    You are a professional scriptwriter for documentary-style voiceovers and cinematic trailers.
-    Given the following scene summaries for an animatic storyboard:
+    voice_style_description = {
+        "cinematic_trailer": "deep, resonant, and slightly epic, building anticipation.",
+        "documentary_neutral": "clear, informative, and objective.",
+        "introspective_character": f"reflective, personal, perhaps echoing the thoughts of a key character (e.g., Jax, if present)."
+    }[voice_style]
 
-    --- SCENE SUMMARIES ---
+    prompt = f"""
+    You are an award-winning voiceover scriptwriter.
+    Craft a compelling, continuous narration script for a short cinematic animatic based on these scene insights:
+    --- SCENE INSIGHTS ---
     {full_summary_text}
-    --- END SCENE SUMMARIES ---
+    --- END SCENE INSIGHTS ---
 
-    Overall Story Genre: {overall_genre}
-    Overall Story Mood: {overall_mood}
+    Overall Genre: {overall_genre}
+    Overall Mood: {overall_mood}
+    Desired Voiceover Style: {voice_style} (Characteristics: {voice_style_description})
 
-    Write a concise, engaging, and continuous narration script that flows smoothly across these scenes.
-    The narration should enhance the visual storytelling, not just describe what's visible.
-    It should set the tone, build suspense or emotion, and connect the scenes thematically.
-    The tone of the narration should match the overall mood and genre.
-    Keep the narration for each scene relatively brief (1-2 short sentences per scene on average).
-    The total narration should be suitable for a short animatic (e.g., if 3 scenes at 4 seconds each, total ~12 seconds of video, so narration should be ~60-90 words max).
-    Do not include scene numbers or explicit directives like "(Voiceover)" in the output. Just provide the pure narration text.
-    Focus on evocative language.
+    The narration should:
+    - Weave a cohesive narrative thread through the scenes.
+    - Enhance emotional impact and atmosphere, guided by the scene's 'emotional_beat' and 'soundscape_hint'.
+    - Be concise: Aim for 1-2 powerful sentences per scene. Total words ~60-100 for 3-4 scenes.
+    - Avoid merely describing the action; instead, offer insight, build tension, or evoke thematic depth.
+    - If style is 'introspective_character', imagine one of the characters is narrating their internal monologue.
 
-    Example (if scenes were about a space discovery):
-    "The red dust of Mars whispered secrets of a forgotten age. Deep within the chasm, an impossible structure pulsed with an alien light, beckoning humanity towards a destiny unknown, and perhaps, a truth too vast to comprehend."
-
-    Output ONLY the narration script text.
+    Output ONLY the pure narration script text, ready for text-to-speech. No scene numbers, no "VO:", just the spoken words.
     """
     return " ".join(prompt.split())
 
-
-# create_scene_regeneration_prompt - REMAINS SAME
+# create_scene_regeneration_prompt - Now uses the new fields
 def create_scene_regeneration_prompt(original_scene_data, user_feedback, full_story_context=None):
-    context_str = f"Original scene (Scene Number {original_scene_data.get('scene_number')}):\n{json.dumps(original_scene_data, indent=2)}\n\n"
+    # ... (context_str same) ...
+    context_str = f"Original scene (Scene Number {original_scene_data.get('scene_number')} - Title: {original_scene_data.get('scene_title')} ):\n{json.dumps(original_scene_data, indent=2)}\n\n"
     if full_story_context: context_str += f"Full story context:\n{json.dumps(full_story_context, indent=2)}\n\n"
-    return (f"Expert script doctor. Original scene:\n{context_str}User feedback: \"{user_feedback}\"\n"
-            f"Regenerate ONLY the JSON for this single scene, incorporating feedback. Maintain structure. 'key_action' max 15-20 words.")
-
-# create_visual_regeneration_prompt - REMAINS SAME
-def create_visual_regeneration_prompt(original_image_prompt_text, user_feedback_on_visuals, scene_data, character_definitions=None, global_style_reference=""):
-    scene_context_summary = (f"Scene: {scene_data.get('emotional_beat', '')}. Setting: {scene_data.get('setting_description', '')}. "
-                             f"Action: {scene_data.get('key_action', '')}. Characters: {', '.join(scene_data.get('characters_involved',[]))}.")
-    char_details_str = "Relevant characters: " + (", ".join([f"{name} ({desc})" for name, desc in character_definitions.items() if name.lower() in [cn.lower() for cn in scene_data.get('characters_involved',[])]])) if character_definitions else "None specified."
-    return (f"AI assistant for refining DALL-E 3 prompts. Original Scene Context: {scene_context_summary} {char_details_str} "
-            f"Global Style: \"{global_style_reference}\". Original DALL-E 3 prompt was: \"{original_image_prompt_text}\". "
-            f"User feedback on visual: \"{user_feedback_on_visuals}\". Generate a new, revised DALL-E 3 prompt. "
-            f"It must be ultra-detailed, photorealistic, cinematic, film/game quality. Translate feedback into concrete visual descriptions. "
-            f"Respect character descriptions. Output ONLY the new prompt string.")
+
+    return f"""
+    You are an AI Script Supervisor and Creative Consultant.
+    {context_str}
+    User Feedback for this scene: "{user_feedback}"
+
+    Regenerate ONLY the JSON object for this single scene, incorporating the feedback.
+    Maintain the exact field structure: (scene_number, scene_title, setting_description, characters_involved, character_focus_moment, key_plot_beat, suggested_dialogue_hook, PROACTIVE_visual_style_감독, PROACTIVE_camera_work_감독, PROACTIVE_sound_design_감독, dalle_image_prompt_keywords, pexels_search_query_감독).
+    'scene_number' MUST NOT change.
+    If feedback targets plot, characters, or dialogue, adjust relevant fields.
+    If feedback targets visuals, camera, or sound, update the 'PROACTIVE_..._감독' fields AND the 'dalle_image_prompt_keywords' and 'pexels_search_query_감독' to reflect the new direction.
+    Ensure 'key_plot_beat' is a concise sentence (max 15-20 words).
+    If adding a `director_note` due to experimental narrative changes, ensure it's brief.
+    """
+
+# create_visual_regeneration_prompt - Now uses more context for Gemini to rewrite DALL-E prompt
+def create_visual_regeneration_prompt(original_dalle_prompt, user_feedback, scene_data, character_definitions=None, global_style_additions=""):
+    # ... (character_narrative same as in construct_dalle_prompt) ...
+    characters_involved_in_scene = scene_data.get('characters_involved', [])
+    character_prompt_segments = []
+    if characters_involved_in_scene:
+        for char_name_from_scene in characters_involved_in_scene:
+            char_name_clean = char_name_from_scene.strip(); char_lookup_key = char_name_clean.lower()
+            if character_definitions and char_lookup_key in character_definitions: character_prompt_segments.append(f"{char_name_clean} ({character_definitions[char_lookup_key]})")
+            else: character_prompt_segments.append(char_name_clean)
+    characters_narrative = f" Characters to feature: {', '.join(character_prompt_segments) if character_prompt_segments else 'None specifically detailed'}."
+
+    full_prompt_for_gemini = f"""
+    You are an AI Art Director specializing in refining DALL-E 3 prompts for cinematic visuals.
+    The goal is to update an image prompt based on user feedback.
+
+    Scene Context:
+    - Title: "{scene_data.get('scene_title', '')}"
+    - Setting: "{scene_data.get('setting_description', '')}"
+    - Key Plot Beat: "{scene_data.get('key_plot_beat', '')}"
+    - {characters_narrative}
+    - Director's Suggested Visual Style: "{scene_data.get('PROACTIVE_visual_style_감독', '')}"
+    - Director's Suggested Camera: "{scene_data.get('PROACTIVE_camera_work_감독', '')}"
+    - Current Global Style Additions: "{global_style_additions}"
+
+    The PREVIOUS DALL-E 3 prompt was:
+    "{original_dalle_prompt}"
+
+    User Feedback on the image generated by the previous prompt:
+    "{user_feedback}"
+
+    Your Task: Generate a NEW, revised DALL-E 3 prompt.
+    This new prompt must incorporate the user's feedback to achieve the desired visual changes.
+    It should remain ultra-detailed, photorealistic, and highly cinematic.
+    The prompt should guide DALL-E 3 to create a stunning image suitable for a film's concept art.
+    Maintain core scene elements (setting, characters, plot beat) unless feedback explicitly requests changes.
+    Translate feedback into concrete visual descriptions (lighting, color, composition, character appearance/pose, atmosphere).
+    Reinforce character descriptions if they are relevant to the feedback.
+
+    Output ONLY the new, revised DALL-E 3 prompt string. Do not add any other text.
+    """
+    return " ".join(full_prompt_for_gemini.split())