CingenAI / app.py
mgbam's picture
Update app.py
a48cea9 verified
raw
history blame
41.3 kB
# app.py
import streamlit as st
from core.gemini_handler import GeminiHandler
from core.visual_engine import VisualEngine
from core.prompt_engineering import (
create_cinematic_treatment_prompt,
construct_dalle_prompt,
construct_text_to_video_prompt_for_gen4, # <<< USE THIS FOR RUNWAY
create_narration_script_prompt_enhanced,
create_scene_regeneration_prompt,
create_visual_regeneration_prompt # This is for DALL-E image prompt refinement
)
import os
import logging
# --- Configuration & Initialization ---
st.set_page_config(page_title="CineGen AI Ultra+", layout="wide", initial_sidebar_state="expanded")
# Configure logging to be more verbose for debugging if needed
logging.basicConfig(level=logging.DEBUG, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__) # Get logger for this module
# --- Global Definitions ---
SHOT_TYPES_OPTIONS = [ "Director's Choice", "Establishing Shot", "Long Shot", "Full Shot", "Medium Long Shot (Cowboy)", "Medium Shot", "Medium Close-up", "Close-up", "Extreme Close-up", "Point of View (POV)", "Over the Shoulder", "Tracking Shot", "Dolly Zoom", "Crane Shot", "Aerial Shot", "Static Shot", "Dutch Angle", "Whip Pan"]
DEFAULT_SCENE_DURATION_SECS = 5
DEFAULT_SHOT_TYPE = "Director's Choice"
ASSET_TYPE_OPTIONS = ["Auto (Director's Choice)", "Image", "Video Clip"]
# --- Global State Variables & API Key Setup ---
def load_api_key(key_name_streamlit, key_name_env, service_name):
# (Keep this function as it was - robust)
key = None; secrets_available = hasattr(st, 'secrets')
try:
if secrets_available and key_name_streamlit in st.secrets:
key = st.secrets[key_name_streamlit]
if key: logger.info(f"{service_name} API Key found in Streamlit secrets.")
except Exception as e: logger.warning(f"Could not access st.secrets for {key_name_streamlit}: {e}")
if not key and key_name_env in os.environ:
key = os.environ[key_name_env]
if key: logger.info(f"{service_name} API Key found in environment variable.")
if not key: logger.warning(f"{service_name} API Key NOT FOUND. Related features may be disabled or use fallbacks.")
return key
if 'services_initialized' not in st.session_state:
logger.info("Initializing services and API keys for the first time this session...")
st.session_state.GEMINI_API_KEY = load_api_key("GEMINI_API_KEY", "GEMINI_API_KEY", "Gemini")
st.session_state.OPENAI_API_KEY = load_api_key("OPENAI_API_KEY", "OPENAI_API_KEY", "OpenAI/DALL-E")
st.session_state.ELEVENLABS_API_KEY = load_api_key("ELEVENLABS_API_KEY", "ELEVENLABS_API_KEY", "ElevenLabs")
st.session_state.PEXELS_API_KEY = load_api_key("PEXELS_API_KEY", "PEXELS_API_KEY", "Pexels")
st.session_state.ELEVENLABS_VOICE_ID_CONFIG = load_api_key("ELEVENLABS_VOICE_ID", "ELEVENLABS_VOICE_ID", "ElevenLabs Voice ID")
st.session_state.RUNWAY_API_KEY = load_api_key("RUNWAY_API_KEY", "RUNWAY_API_KEY", "RunwayML")
if not st.session_state.GEMINI_API_KEY: st.error("CRITICAL: Gemini API Key missing!"); logger.critical("Gemini API Key missing."); st.stop()
try: st.session_state.gemini_handler = GeminiHandler(api_key=st.session_state.GEMINI_API_KEY); logger.info("GeminiHandler initialized.")
except Exception as e: st.error(f"Failed to init GeminiHandler: {e}"); logger.critical(f"GeminiHandler init failed: {e}", exc_info=True); st.stop()
try:
default_voice_id_el = "Rachel" # Fallback
configured_voice_id_el = st.session_state.ELEVENLABS_VOICE_ID_CONFIG or default_voice_id_el
st.session_state.visual_engine = VisualEngine(output_dir="temp_cinegen_media", default_elevenlabs_voice_id=configured_voice_id_el)
st.session_state.visual_engine.set_openai_api_key(st.session_state.OPENAI_API_KEY)
st.session_state.visual_engine.set_elevenlabs_api_key(st.session_state.ELEVENLABS_API_KEY, voice_id_from_secret=st.session_state.ELEVENLABS_VOICE_ID_CONFIG)
st.session_state.visual_engine.set_pexels_api_key(st.session_state.PEXELS_API_KEY)
st.session_state.visual_engine.set_runway_api_key(st.session_state.RUNWAY_API_KEY) # Pass Runway key
logger.info("VisualEngine initialized and API keys set.")
except Exception as e: st.error(f"Failed to init VisualEngine: {e}"); logger.critical(f"VisualEngine init/key setting failed: {e}", exc_info=True); st.warning("VisualEngine critical setup issue.")
st.session_state.services_initialized = True; logger.info("Service initialization complete.")
# Initialize other session state variables
for key_ss, default_val_ss in [ # Renamed loop vars
('story_treatment_scenes', []), ('scene_generation_prompts', []), ('generated_scene_assets_info', []), # Stores full asset info dicts
('video_path', None), ('character_definitions', {}), ('global_style_additions', ""),
('overall_narration_audio_path', None), ('narration_script_display', "")
]:
if key_ss not in st.session_state: st.session_state[key_ss] = default_val_ss
def initialize_new_project_state(): # Renamed
st.session_state.story_treatment_scenes = []
st.session_state.scene_generation_prompts = [] # Stores the prompt used for DALL-E or Runway
st.session_state.generated_scene_assets_info = [] # Stores dicts {'path':..., 'type':..., 'error':..., 'prompt_used':...}
st.session_state.video_path, st.session_state.overall_narration_audio_path, st.session_state.narration_script_display = None, None, ""
logger.info("New project state initialized.")
def generate_asset_for_scene_wrapper(scene_idx, scene_dict_data, version_num=1, user_selected_asset_type_override="Auto (Director's Choice)"): # Renamed
# Determine if video clip is desired based on user override or Gemini's suggestion
generate_as_video_clip_final = False
gemini_suggested_asset_type = scene_dict_data.get('suggested_asset_type_감독', 'image').lower()
if user_selected_asset_type_override == "Image":
generate_as_video_clip_final = False
elif user_selected_asset_type_override == "Video Clip":
generate_as_video_clip_final = True
elif user_selected_asset_type_override == "Auto (Director's Choice)": # Default
generate_as_video_clip_final = (gemini_suggested_asset_type == "video_clip")
# Prompt for base image generation (DALL-E or Pexels fallback)
image_gen_prompt_text = construct_dalle_prompt(scene_dict_data, st.session_state.character_definitions, st.session_state.global_style_additions)
# Prompt for video motion (Runway Gen-4) - only if generating video
motion_gen_prompt_text = ""
if generate_as_video_clip_final:
motion_gen_prompt_text = construct_text_to_video_prompt_for_gen4(scene_dict_data, st.session_state.global_style_additions)
if not motion_gen_prompt_text: # Fallback if specific motion prompt is empty
logger.warning(f"S{scene_dict_data.get('scene_number', scene_idx+1)}: Motion prompt empty, using generic for Runway.")
motion_gen_prompt_text = scene_dict_data.get('video_clip_motion_description_감독', "subtle ambient motion")
if not image_gen_prompt_text: # Base image prompt is always needed
logger.error(f"Base image prompt construction failed for S{scene_dict_data.get('scene_number', scene_idx+1)}"); return False
# Ensure session state lists are adequate
while len(st.session_state.scene_generation_prompts) <= scene_idx: st.session_state.scene_generation_prompts.append("")
while len(st.session_state.generated_scene_assets_info) <= scene_idx: st.session_state.generated_scene_assets_info.append(None)
# Store the relevant prompt (DALL-E for image, motion for video)
# The generate_scene_asset method will return the actual prompt it used if different internally.
st.session_state.scene_generation_prompts[scene_idx] = motion_gen_prompt_text if generate_as_video_clip_final else image_gen_prompt_text
filename_base_for_asset = f"scene_{scene_dict_data.get('scene_number', scene_idx+1)}_asset_v{version_num}" # Renamed
runway_dur_for_scene = scene_dict_data.get('video_clip_duration_estimate_secs_감독', scene_dict_data.get('user_scene_duration_secs', DEFAULT_SCENE_DURATION_SECS))
if runway_dur_for_scene <= 0 : runway_dur_for_scene = DEFAULT_SCENE_DURATION_SECS
asset_result_dict = st.session_state.visual_engine.generate_scene_asset(
image_generation_prompt_text=image_gen_prompt_text, # For base DALL-E/Pexels image
motion_prompt_text_for_video=motion_gen_prompt_text, # For Runway motion
scene_data=scene_dict_data,
scene_identifier_filename_base=filename_base_for_asset,
generate_as_video_clip=generate_as_video_clip_final,
runway_target_duration=runway_dur_for_scene
)
st.session_state.generated_scene_assets_info[scene_idx] = asset_result_dict
# Update the stored prompt with what was actually used by the engine, if available from result
if asset_result_dict and asset_result_dict.get('prompt_used'):
st.session_state.scene_generation_prompts[scene_idx] = asset_result_dict['prompt_used']
if asset_result_dict and not asset_result_dict['error'] and asset_result_dict.get('path') and os.path.exists(asset_result_dict['path']):
logger.info(f"Asset ({asset_result_dict.get('type')}) generated for S{scene_dict_data.get('scene_number', scene_idx+1)}: {os.path.basename(asset_result_dict['path'])}")
return True
else:
err_msg_asset = asset_result_dict.get('error_message', 'Unknown error') if asset_result_dict else 'Asset result is None'
logger.warning(f"Asset gen FAILED for S{scene_dict_data.get('scene_number', scene_idx+1)}. Type attempted: {'Video' if generate_as_video_clip_final else 'Image'}. Error: {err_msg_asset}")
# Store a more detailed failure state if not already
if not st.session_state.generated_scene_assets_info[scene_idx] or not st.session_state.generated_scene_assets_info[scene_idx]['error']:
st.session_state.generated_scene_assets_info[scene_idx] = {'path': None, 'type': 'none', 'error': True, 'error_message': err_msg_asset, 'prompt_used': st.session_state.scene_generation_prompts[scene_idx]}
return False
# --- UI Sidebar ---
with st.sidebar:
# ... (Sidebar UI code as before, no changes needed for this fix) ...
st.title("🎬 CineGen AI Ultra+")
st.markdown("### Creative Seed")
user_idea = st.text_area("Core Story Idea / Theme:", "A lone wanderer searches for a mythical oasis in a vast, post-apocalyptic desert, haunted by mirages and mechanical scavengers.", height=120, key="user_idea_main_v5")
genre = st.selectbox("Primary Genre:", ["Cyberpunk", "Sci-Fi", "Fantasy", "Noir", "Thriller", "Western", "Post-Apocalyptic", "Historical Drama", "Surreal"], index=6, key="genre_main_v5")
mood = st.selectbox("Overall Mood:", ["Hopeful yet Desperate", "Mysterious & Eerie", "Gritty & Tense", "Epic & Awe-Inspiring", "Melancholy & Reflective", "Whimsical & Lighthearted"], index=0, key="mood_main_v5")
num_scenes = st.slider("Number of Key Scenes:", 1, 10, 2, key="num_scenes_main_v5")
creative_guidance_options = {"Standard Director": "standard", "Artistic Visionary": "more_artistic", "Experimental Storyteller": "experimental_narrative"}
selected_creative_guidance_key = st.selectbox("AI Creative Director Style:", options=list(creative_guidance_options.keys()), key="creative_guidance_select_v5")
actual_creative_guidance = creative_guidance_options[selected_creative_guidance_key]
if st.button("🌌 Generate Cinematic Treatment", type="primary", key="generate_treatment_btn_v5", use_container_width=True):
initialize_new_project_state() # Use renamed function
if not user_idea.strip(): st.warning("Please provide a story idea.")
else:
with st.status("AI Director is envisioning your masterpiece...", expanded=True) as status_op: # Renamed
try:
status_op.write("Phase 1: Gemini crafting cinematic treatment... πŸ“œ"); logger.info("Phase 1: Cinematic Treatment Gen.")
treatment_gen_prompt = create_cinematic_treatment_prompt(user_idea, genre, mood, num_scenes, actual_creative_guidance) # Renamed
raw_treatment_result = st.session_state.gemini_handler.generate_story_breakdown(treatment_gen_prompt) # Renamed
if not isinstance(raw_treatment_result, list) or not raw_treatment_result: raise ValueError("Gemini returned invalid scene list format.")
processed_scene_list = [] # Renamed
for scene_from_gemini in raw_treatment_result: # Renamed
scene_from_gemini['user_shot_type'] = scene_from_gemini.get('PROACTIVE_camera_work_감독', DEFAULT_SHOT_TYPE)
# Use Gemini's video duration estimate if available for video clips, else default scene duration
gemini_dur_est = scene_from_gemini.get('video_clip_duration_estimate_secs_감독', 0)
scene_from_gemini['user_scene_duration_secs'] = gemini_dur_est if gemini_dur_est > 0 else DEFAULT_SCENE_DURATION_SECS
scene_from_gemini['user_selected_asset_type'] = "Auto (Director's Choice)" # UI default
processed_scene_list.append(scene_from_gemini)
st.session_state.story_treatment_scenes = processed_scene_list
num_generated_scenes = len(st.session_state.story_treatment_scenes) # Renamed
st.session_state.scene_generation_prompts = [""]*num_generated_scenes
st.session_state.generated_scene_assets_info = [None]*num_generated_scenes
logger.info(f"Phase 1 complete. {num_generated_scenes} scenes."); status_op.update(label="Treatment complete! βœ… Generating visuals...", state="running")
status_op.write("Phase 2: Creating visual assets (Image/Video)... πŸ–ΌοΈπŸŽ¬"); logger.info("Phase 2: Visual Asset Gen.")
successful_asset_count = 0 # Renamed
for i_scene, scene_data_item in enumerate(st.session_state.story_treatment_scenes): # Renamed
scene_num_display = scene_data_item.get('scene_number', i_scene+1) # Renamed
status_op.write(f" Asset for Scene {scene_num_display}..."); logger.info(f" Processing asset for Scene {scene_num_display}.")
if generate_asset_for_scene_wrapper(i_scene, scene_data_item, version_num=1): # Pass default 'Auto' for initial gen
successful_asset_count += 1
status_label_phase2 = "Visual assets ready! " # Renamed
next_op_state = "running" # Renamed
if successful_asset_count == 0 and num_generated_scenes > 0:
logger.error("Asset gen failed for all scenes."); status_label_phase2 = "Asset gen FAILED for all scenes."; next_op_state="error";
status_op.update(label=status_label_phase2, state=next_op_state, expanded=True); st.stop()
elif successful_asset_count < num_generated_scenes:
logger.warning(f"Assets partially generated ({successful_asset_count}/{num_generated_scenes})."); status_label_phase2 = f"Assets partially generated ({successful_asset_count}/{num_generated_scenes}). "
status_op.update(label=f"{status_label_phase2}Generating narration script...", state=next_op_state)
if next_op_state == "error": st.stop()
status_op.write("Phase 3: Generating narration script..."); logger.info("Phase 3: Narration Script Gen.")
voice_style_for_narration_prompt = st.session_state.get("selected_voice_style_for_generation", "cinematic_trailer") # Renamed
narration_gen_prompt = create_narration_script_prompt_enhanced(st.session_state.story_treatment_scenes, mood, genre, voice_style_for_narration_prompt) # Renamed
st.session_state.narration_script_display = st.session_state.gemini_handler.generate_image_prompt(narration_gen_prompt) # generate_image_prompt returns string
logger.info("Narration script generated."); status_op.update(label="Narration script ready! Synthesizing voice...", state="running")
status_op.write("Phase 4: Synthesizing voice (ElevenLabs)... πŸ”Š"); logger.info("Phase 4: Voice Synthesis.")
st.session_state.overall_narration_audio_path = st.session_state.visual_engine.generate_narration_audio(st.session_state.narration_script_display)
final_status_label = "All components ready! Storyboard below. πŸš€" # Renamed
final_op_state = "complete" # Renamed
if not st.session_state.overall_narration_audio_path:
final_status_label = f"{status_label_phase2}Storyboard ready (Voiceover skipped or failed)."
logger.warning("Voiceover generation was skipped or failed.")
else: logger.info("Voiceover generated successfully.")
status_op.update(label=final_status_label, state=final_op_state, expanded=False)
except ValueError as ve_err: logger.error(f"ValueError in main generation: {ve_err}", exc_info=True); status_op.update(label=f"Input or Gemini response error: {ve_err}", state="error", expanded=True); # Renamed
except Exception as e_unhandled: logger.error(f"Unhandled Exception in main generation: {e_unhandled}", exc_info=True); status_op.update(label=f"An unexpected error: {e_unhandled}", state="error", expanded=True); # Renamed
# --- Sidebar Fine-Tuning Options (Characters, Global Style, Voice) ---
# (Keep these sections as they were in the previous correct version)
with st.expander("Define Characters", expanded=False):
char_name_input = st.text_input("Character Name", key="char_name_adv_ultra_v5_sb"); char_desc_input = st.text_area("Visual Description", key="char_desc_adv_ultra_v5_sb", height=100, placeholder="e.g., Jax: rugged male astronaut...")
if st.button("Save Character", key="add_char_adv_ultra_v5_sb"):
if char_name_input and char_desc_input: st.session_state.character_definitions[char_name_input.strip().lower()] = char_desc_input.strip(); st.success(f"Char '{char_name_input.strip()}' saved.")
else: st.warning("Name and description needed.")
if st.session_state.character_definitions: st.caption("Current Characters:"); [st.markdown(f"**{k.title()}:** _{v}_") for k,v in st.session_state.character_definitions.items()]
with st.expander("Global Style Overrides", expanded=False):
style_presets_dict = { "Default (Director's Choice)": "", "Hyper-Realistic Gritty Noir": "hyper-realistic gritty neo-noir...", "Surreal Dreamscape Fantasy": "surreal dreamscape, epic fantasy...", "Vintage Analog Sci-Fi": "70s/80s analog sci-fi..."} # Truncated for brevity
selected_style_preset_key = st.selectbox("Base Style Preset:", options=list(style_presets_dict.keys()), key="style_preset_adv_ultra_v5_sb")
custom_style_keywords_input = st.text_area("Additional Custom Style Keywords:", key="custom_style_adv_ultra_v5_sb", height=80, placeholder="e.g., 'Dutch angle'")
current_global_style = st.session_state.global_style_additions
if st.button("Apply Global Styles", key="apply_styles_adv_ultra_v5_sb"):
final_style_str = style_presets_dict[selected_style_preset_key];
if custom_style_keywords_input.strip(): final_style_str = f"{final_style_str}, {custom_style_keywords_input.strip()}" if final_style_str else custom_style_keywords_input.strip()
st.session_state.global_style_additions = final_style_str.strip(); current_global_style = final_style_str.strip() # Update local var for immediate display
if current_global_style: st.success("Global styles applied!")
else: st.info("Global style additions cleared.")
if current_global_style: st.caption(f"Active global styles: \"{current_global_style}\"")
with st.expander("Voice & Narration Style", expanded=False):
engine_default_voice = "Rachel"
if hasattr(st.session_state, 'visual_engine') and st.session_state.visual_engine: engine_default_voice = st.session_state.visual_engine.elevenlabs_voice_id
user_voice_id_input = st.text_input("ElevenLabs Voice ID (override):", value=engine_default_voice, key="el_voice_id_override_v5_sb", help=f"Defaulting to '{engine_default_voice}'.")
narration_prompt_styles_dict = {"Cinematic Trailer": "cinematic_trailer", "Neutral Documentary": "documentary_neutral", "Character Introspection": "introspective_character"}
selected_narration_style_key = st.selectbox("Narration Script Style:", list(narration_prompt_styles_dict.keys()), key="narr_style_sel_v5_sb", index=0)
if st.button("Set Narrator Voice & Style", key="set_voice_btn_ultra_v5_sb"):
final_voice_id_to_use_el = user_voice_id_input.strip() or st.session_state.get("ELEVENLABS_VOICE_ID_CONFIG", "Rachel") # Fallback
if hasattr(st.session_state, 'visual_engine'): st.session_state.visual_engine.elevenlabs_voice_id = final_voice_id_to_use_el
st.session_state.selected_voice_style_for_generation = narration_prompt_styles_dict[selected_narration_style_key]
st.success(f"Narrator Voice ID: {final_voice_id_to_use_el}. Script Style: {selected_narration_style_key}")
logger.info(f"User updated 11L Voice ID: {final_voice_id_to_use_el}, Script Style: {selected_narration_style_key}")
# --- Main Content Area ---
st.header("🎬 Cinematic Storyboard & Treatment")
if st.session_state.narration_script_display:
with st.expander("πŸ“œ View Full Narration Script", expanded=False): st.markdown(f"> _{st.session_state.narration_script_display}_")
if not st.session_state.story_treatment_scenes: st.info("Use the sidebar to generate your cinematic treatment.")
else:
for i_main_loop, scene_content_item in enumerate(st.session_state.story_treatment_scenes): # Renamed
scene_num_val = scene_content_item.get('scene_number', i_main_loop + 1) # Renamed
scene_title_val = scene_content_item.get('scene_title', 'Untitled Scene') # Renamed
# Ensure unique keys for widgets within the loop
key_base_for_scene = f"s{scene_num_val}_{''.join(filter(str.isalnum, scene_title_val[:10]))}_main_{i_main_loop}" # Renamed
if "director_note" in scene_content_item and scene_content_item['director_note']: st.info(f"🎬 Director Note S{scene_num_val}: {scene_content_item['director_note']}")
st.subheader(f"SCENE {scene_num_val}: {scene_title_val.upper()}"); col_treatment, col_visual = st.columns([0.45, 0.55]) # Renamed
with col_treatment: # Treatment and Controls Column
with st.expander("πŸ“ Scene Treatment & Controls", expanded=True):
# ... (Display textual scene details - beat, setting, chars, etc. - as before) ...
st.markdown(f"**Beat:** {scene_content_item.get('emotional_beat', 'N/A')}"); st.markdown(f"**Setting:** {scene_content_item.get('setting_description', 'N/A')}"); st.markdown(f"**Chars:** {', '.join(scene_content_item.get('characters_involved', ['N/A']))}"); st.markdown(f"**Focus Moment:** _{scene_content_item.get('character_focus_moment', 'N/A')}_"); st.markdown(f"**Plot Beat:** {scene_content_item.get('key_plot_beat', 'N/A')}"); st.markdown(f"**Dialogue Hook:** `\"{scene_content_item.get('suggested_dialogue_hook', '...')}\"`"); st.markdown("---"); st.markdown(f"**Dir. Visual Style:** _{scene_content_item.get('PROACTIVE_visual_style_감독', 'N/A')}_"); st.markdown(f"**Dir. Camera:** _{scene_content_item.get('PROACTIVE_camera_work_감독', 'N/A')}_"); st.markdown(f"**Dir. Sound:** _{scene_content_item.get('PROACTIVE_sound_design_감독', 'N/A')}_"); st.markdown("---")
st.markdown("##### Shot, Pacing & Asset Controls")
# User Shot Type (Camera Angle)
current_ui_shot_type = st.session_state.story_treatment_scenes[i_main_loop].get('user_shot_type', DEFAULT_SHOT_TYPE) # Renamed
try: shot_type_idx_val = SHOT_TYPES_OPTIONS.index(current_ui_shot_type) # Renamed
except ValueError: shot_type_idx_val = SHOT_TYPES_OPTIONS.index(DEFAULT_SHOT_TYPE)
new_ui_shot_type = st.selectbox("Dominant Shot Type:", options=SHOT_TYPES_OPTIONS, index=shot_type_idx_val, key=f"shot_type_widget_{key_base_for_scene}") # Renamed
if new_ui_shot_type != current_ui_shot_type: st.session_state.story_treatment_scenes[i_main_loop]['user_shot_type'] = new_ui_shot_type
# User Scene Duration
current_ui_duration = st.session_state.story_treatment_scenes[i_main_loop].get('user_scene_duration_secs', DEFAULT_SCENE_DURATION_SECS) # Renamed
new_ui_duration = st.number_input("Scene Duration (seconds):", min_value=1, max_value=300, value=current_ui_duration, step=1, key=f"duration_widget_{key_base_for_scene}") # Renamed
if new_ui_duration != current_ui_duration: st.session_state.story_treatment_scenes[i_main_loop]['user_scene_duration_secs'] = new_ui_duration
# User Asset Type Selection
current_ui_asset_type = st.session_state.story_treatment_scenes[i_main_loop].get('user_selected_asset_type', "Auto (Director's Choice)") # Renamed
try: asset_type_idx_val = ASSET_TYPE_OPTIONS.index(current_ui_asset_type) # Renamed
except ValueError: asset_type_idx_val = 0
new_ui_asset_type = st.selectbox("Asset Type Override:", ASSET_TYPE_OPTIONS, index=asset_type_idx_val, key=f"asset_type_sel_{key_base_for_scene}", help="Choose 'Image' or 'Video Clip'. 'Auto' uses Gemini's suggestion.") # Renamed
if new_ui_asset_type != current_ui_asset_type: st.session_state.story_treatment_scenes[i_main_loop]['user_selected_asset_type'] = new_ui_asset_type
st.markdown("---")
# Display generated prompt for the asset
prompt_for_current_asset = st.session_state.scene_generation_prompts[i_main_loop] if i_main_loop < len(st.session_state.scene_generation_prompts) else None # Renamed
if prompt_for_current_asset:
with st.popover("πŸ‘οΈ View Asset Generation Prompt"):
st.markdown(f"**Prompt used for current asset:**"); st.code(prompt_for_current_asset, language='text')
pexels_query_val = scene_content_item.get('pexels_search_query_감독', None) # Renamed
if pexels_query_val: st.caption(f"Pexels Fallback Query: `{pexels_query_val}`")
with col_visual: # Visuals Column
asset_info_for_scene = st.session_state.generated_scene_assets_info[i_main_loop] if i_main_loop < len(st.session_state.generated_scene_assets_info) else None # Renamed
if asset_info_for_scene and not asset_info_for_scene.get('error') and asset_info_for_scene.get('path') and os.path.exists(asset_info_for_scene['path']):
path_to_asset_file = asset_info_for_scene['path'] # Renamed
type_of_asset_file = asset_info_for_scene.get('type', 'image') # Renamed
if type_of_asset_file == 'image': st.image(path_to_asset_file, caption=f"Scene {scene_num_val} ({type_of_asset_file}): {scene_title_val}")
elif type_of_asset_file == 'video':
try:
with open(path_to_asset_file, 'rb') as vf_read: video_bytes_data = vf_read.read() # Renamed
st.video(video_bytes_data, format="video/mp4", start_time=0); st.caption(f"Scene {scene_num_val} ({type_of_asset_file}): {scene_title_val}")
except Exception as e_vid_display: st.error(f"Error displaying video {path_to_asset_file}: {e_vid_display}"); logger.error(f"Error displaying video: {e_vid_display}", exc_info=True) # Renamed
else: st.warning(f"Unknown asset type '{type_of_asset_file}' for Scene {scene_num_val}.")
else:
if st.session_state.story_treatment_scenes:
error_message_display = asset_info_for_scene.get('error_message', 'Visual pending or failed.') if asset_info_for_scene else 'Visual pending or failed.' # Renamed
st.caption(error_message_display)
# --- Popovers for Editing Scene Treatment & Visual Prompt ---
with st.popover(f"✏️ Edit S{scene_num_val} Treatment"):
feedback_for_treatment = st.text_area("Changes to treatment:", key=f"treat_fb_{key_base_for_scene}", height=150) # Renamed
if st.button(f"πŸ”„ Update S{scene_num_val} Treatment", key=f"regen_treat_btn_{key_base_for_scene}"):
if feedback_for_treatment:
with st.status(f"Updating S{scene_num_val} Treatment & Asset...", expanded=True) as status_treatment_regen: # Renamed
user_shot_type_pref = st.session_state.story_treatment_scenes[i_main_loop]['user_shot_type'] # Renamed
user_duration_pref = st.session_state.story_treatment_scenes[i_main_loop]['user_scene_duration_secs'] # Renamed
user_asset_type_pref = st.session_state.story_treatment_scenes[i_main_loop]['user_selected_asset_type'] # Renamed
regen_prompt_for_gemini = create_scene_regeneration_prompt(scene_content_item, feedback_for_treatment, st.session_state.story_treatment_scenes) # Renamed
try:
updated_scene_data_gemini = st.session_state.gemini_handler.regenerate_scene_script_details(regen_prompt_for_gemini) # Renamed
final_updated_scene_data = {**updated_scene_data_gemini} # Renamed
final_updated_scene_data['user_shot_type'] = user_shot_type_pref
final_updated_scene_data['user_scene_duration_secs'] = user_duration_pref
final_updated_scene_data['user_selected_asset_type'] = user_asset_type_pref
st.session_state.story_treatment_scenes[i_main_loop] = final_updated_scene_data
status_treatment_regen.update(label="Treatment updated! Regenerating asset...", state="running")
version_num_asset = 1 # Renamed
if asset_info_for_scene and asset_info_for_scene.get('path') and os.path.exists(asset_info_for_scene['path']):
try: base_fn_asset,_=os.path.splitext(os.path.basename(asset_info_for_scene['path'])); version_num_asset = int(base_fn_asset.split('_v')[-1])+1 if '_v' in base_fn_asset else 2 # Renamed
except: version_num_asset = 2
if generate_asset_for_scene_wrapper(i_main_loop, final_updated_scene_data, version_num=version_num_asset, user_selected_asset_type_override=user_asset_type_pref):
status_treatment_regen.update(label="Treatment & Asset Updated! πŸŽ‰", state="complete", expanded=False)
else: status_treatment_regen.update(label="Treatment updated, asset failed.", state="complete", expanded=False)
st.rerun()
except Exception as e_treat_regen_main: status_treatment_regen.update(label=f"Error: {e_treat_regen_main}", state="error"); logger.error(f"Scene treatment regen error: {e_treat_regen_main}", exc_info=True) # Renamed
else: st.warning("Please provide feedback for treatment.")
with st.popover(f"🎨 Edit S{scene_num_val} Visual Prompt/Asset"):
current_gen_prompt_display = st.session_state.scene_generation_prompts[i_main_loop] if i_main_loop < len(st.session_state.scene_generation_prompts) else "No prompt generated yet." # Renamed
st.caption("Current Asset Generation Prompt:"); st.code(current_gen_prompt_display, language='text')
feedback_for_visual = st.text_area("Describe changes for the visual asset:", key=f"visual_fb_{key_base_for_scene}", height=150) # Renamed
if st.button(f"πŸ”„ Update S{scene_num_val} Asset", key=f"regen_visual_btn_{key_base_for_scene}"):
if feedback_for_visual:
with st.status(f"Refining prompt & regenerating asset for S{scene_num_val}...", expanded=True) as status_visual_regen: # Renamed
user_asset_type_choice_visual = st.session_state.story_treatment_scenes[i_main_loop]['user_selected_asset_type'] # Renamed
is_video_asset_type = (user_asset_type_choice_visual == "Video Clip") or \
(user_asset_type_choice_visual == "Auto (Director's Choice)" and scene_content_item.get('suggested_asset_type_감독') == 'video_clip')
newly_constructed_asset_prompt = "" # Renamed
if not is_video_asset_type: # Refining an IMAGE prompt
gemini_refinement_prompt = create_visual_regeneration_prompt(current_gen_prompt_display, feedback_for_visual, scene_content_item, st.session_state.character_definitions, st.session_state.global_style_additions) # Renamed
try:
newly_constructed_asset_prompt = st.session_state.gemini_handler.refine_image_prompt_from_feedback(gemini_refinement_prompt)
st.session_state.scene_generation_prompts[i_main_loop] = newly_constructed_asset_prompt
status_visual_regen.update(label="Image prompt refined by Gemini! Regenerating asset...", state="running")
except Exception as e_gemini_prompt_refine: status_visual_regen.update(label=f"Error refining prompt: {e_gemini_prompt_refine}", state="error"); logger.error(f"Visual prompt refinement error: {e_gemini_prompt_refine}", exc_info=True); continue # Skip asset gen
else: # For VIDEO, reconstruct the motion prompt based on current scene data and feedback (feedback isn't directly used by construct_text_to_video_prompt_for_gen4 here, but scene_data might have changed)
# For video, feedback should ideally modify scene_content_item's motion description first, then reconstruct.
# Simple reconstruction for now:
logger.info(f"Reconstructing video motion prompt for S{scene_num_val} based on feedback (indirectly via scene_data). Feedback was: {feedback_for_visual}")
newly_constructed_asset_prompt = construct_text_to_video_prompt_for_gen4(scene_content_item, st.session_state.global_style_additions)
st.session_state.scene_generation_prompts[i_main_loop] = newly_constructed_asset_prompt
status_visual_regen.update(label="Video motion prompt reconstructed! Regenerating asset...", state="running")
if not newly_constructed_asset_prompt: status_visual_regen.update(label="Prompt construction failed.", state="error"); continue
version_num_visual_asset = 1 # Renamed
if asset_info_for_scene and asset_info_for_scene.get('path') and os.path.exists(asset_info_for_scene['path']):
try: base_fn_viz_asset,_=os.path.splitext(os.path.basename(asset_info_for_scene['path'])); version_num_visual_asset = int(base_fn_viz_asset.split('_v')[-1])+1 if '_v' in base_fn_viz_asset else 2 # Renamed
except: version_num_visual_asset = 2
if generate_asset_for_scene_wrapper(i_main_loop, st.session_state.story_treatment_scenes[i_main_loop], version_num=version_num_visual_asset, user_selected_asset_type_override=user_asset_type_choice_visual):
status_visual_regen.update(label="Asset Updated! πŸŽ‰", state="complete", expanded=False)
else: status_visual_regen.update(label="Prompt updated, asset regeneration failed.", state="complete", expanded=False)
st.rerun()
else: st.warning("Please provide feedback for visual asset.")
st.markdown("---")
# Video Assembly Button
if st.session_state.story_treatment_scenes and any(asset_info_item_loop and not asset_info_item_loop.get('error') and asset_info_item_loop.get('path') for asset_info_item_loop in st.session_state.generated_scene_assets_info if asset_info_item_loop is not None):
if st.button("🎬 Assemble Narrated Cinematic Animatic", key="assemble_ultra_video_btn_v5_main", type="primary", use_container_width=True): # Unique key
with st.status("Assembling Ultra Animatic...", expanded=True) as status_video_assembly: # Renamed
assets_for_final_video = [] # Renamed
for i_vid_assembly, scene_data_for_vid in enumerate(st.session_state.story_treatment_scenes): # Renamed
asset_info_current_scene = st.session_state.generated_scene_assets_info[i_vid_assembly] if i_vid_assembly < len(st.session_state.generated_scene_assets_info) else None # Renamed
if asset_info_current_scene and not asset_info_current_scene.get('error') and asset_info_current_scene.get('path') and os.path.exists(asset_info_current_scene['path']):
assets_for_final_video.append({
'path': asset_info_current_scene['path'],
'type': asset_info_current_scene.get('type', 'image'),
'scene_num': scene_data_for_vid.get('scene_number', i_vid_assembly + 1),
'key_action': scene_data_for_vid.get('key_plot_beat', ''),
'duration': scene_data_for_vid.get('user_scene_duration_secs', DEFAULT_SCENE_DURATION_SECS)
})
status_video_assembly.write(f"Adding S{scene_data_for_vid.get('scene_number', i_vid_assembly + 1)} ({asset_info_current_scene.get('type')}).")
else: logger.warning(f"Skipping S{scene_data_for_vid.get('scene_number', i_vid_assembly+1)} for video: No valid asset.")
if assets_for_final_video:
status_video_assembly.write("Calling video engine...");
st.session_state.video_path = st.session_state.visual_engine.assemble_animatic_from_assets(
asset_data_list=assets_for_final_video,
overall_narration_path=st.session_state.overall_narration_audio_path,
output_filename="cinegen_ultra_animatic.mp4", fps=24
)
if st.session_state.video_path and os.path.exists(st.session_state.video_path):
status_video_assembly.update(label="Ultra animatic assembled! πŸŽ‰", state="complete", expanded=False); st.balloons()
else: status_video_assembly.update(label="Video assembly failed. Check logs.", state="error", expanded=False); logger.error("Video assembly returned None or file does not exist.")
else: status_video_assembly.update(label="No valid assets for video assembly.", state="error", expanded=False); logger.warning("No valid assets found for video assembly.")
elif st.session_state.story_treatment_scenes: st.info("Generate visual assets before assembling the animatic.")
if st.session_state.video_path and os.path.exists(st.session_state.video_path):
st.header("🎬 Generated Cinematic Animatic");
try:
with open(st.session_state.video_path, 'rb') as vf_obj_read: video_bytes_content = vf_obj_read.read() # Renamed
st.video(video_bytes_content, format="video/mp4")
st.download_button(label="Download Ultra Animatic", data=video_bytes_content, file_name=os.path.basename(st.session_state.video_path), mime="video/mp4", use_container_width=True, key="download_ultra_video_btn_v5_main_dl" ) # Unique key
except Exception as e_vid_final_display: st.error(f"Error displaying final video: {e_vid_final_display}"); logger.error(f"Error displaying final video: {e_vid_final_display}", exc_info=True) # Renamed
# --- Footer ---
st.sidebar.markdown("---"); st.sidebar.caption("CineGen AI Ultra+ | Visionary Cinematic Pre-Production")