CingenAI / app.py
mgbam's picture
Update app.py
1813b8c verified
raw
history blame
40.7 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, # Import new function
create_narration_script_prompt_enhanced,
create_scene_regeneration_prompt,
create_visual_regeneration_prompt
)
import os
import logging
# --- Configuration & Initialization ---
st.set_page_config(page_title="CineGen AI Ultra+", layout="wide", initial_sidebar_state="expanded")
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)
# --- Global Definitions for New Features ---
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"] # For user selection
# --- Global State Variables & API Key Setup ---
def load_api_key(key_name_streamlit, key_name_env, service_name):
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") # Load Runway Key
if not st.session_state.GEMINI_API_KEY:
st.error("CRITICAL: Gemini API Key is essential and missing!"); logger.critical("Gemini API Key missing. Halting."); st.stop()
try:
st.session_state.gemini_handler = GeminiHandler(api_key=st.session_state.GEMINI_API_KEY)
logger.info("GeminiHandler initialized successfully.")
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 = "Rachel"
configured_voice_id = st.session_state.ELEVENLABS_VOICE_ID_CONFIG or default_voice_id
st.session_state.visual_engine = VisualEngine(
output_dir="temp_cinegen_media",
default_elevenlabs_voice_id=configured_voice_id
)
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) # Set Runway Key
logger.info("VisualEngine initialized and API keys set (or attempted).")
except Exception as e:
st.error(f"Failed to init VisualEngine or set its API keys: {e}"); logger.critical(f"VisualEngine init/key setting failed: {e}", exc_info=True)
st.warning("VisualEngine critical setup issue. Some features will be disabled.")
st.session_state.services_initialized = True; logger.info("Service initialization sequence complete.")
# Initialize other session state variables
# <<< MODIFIED START >>> : Renamed generated_visual_paths to generated_scene_assets
for key, default_val in [
('story_treatment_scenes', []), ('scene_prompts', []), ('generated_scene_assets', []), # Stores dicts: {'path':..., 'type':...}
('video_path', None), ('character_definitions', {}), ('global_style_additions', ""),
('overall_narration_audio_path', None), ('narration_script_display', "")
]:
if key not in st.session_state: st.session_state[key] = default_val
def initialize_new_project():
st.session_state.story_treatment_scenes = []
st.session_state.scene_prompts = [] # Stores DALL-E or Text-to-Video prompts
st.session_state.generated_scene_assets = [] # Stores dicts {'path': ..., 'type': ..., 'error': ...}
st.session_state.video_path, st.session_state.overall_narration_audio_path, st.session_state.narration_script_display = None, None, ""
logger.info("New project initialized.")
# <<< MODIFIED END >>>
# <<< MODIFIED START >>> : Updated function to use generate_scene_asset
def generate_asset_for_scene_core(scene_index, scene_data, version=1, user_selected_asset_type="Auto (Director's Choice)"):
"""
Generates a visual asset (image or video clip) for a scene.
Returns True on success, False on failure.
"""
# Determine asset type: user override > Gemini suggestion > default to image
final_asset_type_decision = "image" # Default
gemini_suggested_type = scene_data.get('suggested_asset_type_감독', 'image').lower()
if user_selected_asset_type == "Image":
final_asset_type_decision = "image"
elif user_selected_asset_type == "Video Clip":
final_asset_type_decision = "video_clip"
elif user_selected_asset_type == "Auto (Director's Choice)":
final_asset_type_decision = gemini_suggested_type if gemini_suggested_type == "video_clip" else "image"
generate_as_video = (final_asset_type_decision == "video_clip")
prompt_text_for_visual = ""
if generate_as_video:
# Construct prompt for text-to-video (e.g., RunwayML)
prompt_text_for_visual = construct_text_to_video_prompt(scene_data, st.session_state.character_definitions, st.session_state.global_style_additions)
# Note: seed_image_path could be an enhancement if DALL-E image is generated first
else:
# Construct prompt for DALL-E (image)
prompt_text_for_visual = construct_dalle_prompt(scene_data, st.session_state.character_definitions, st.session_state.global_style_additions)
if not prompt_text_for_visual:
logger.error(f"Visual prompt construction failed for scene {scene_data.get('scene_number', scene_index+1)} (Type: {final_asset_type_decision})")
return False
# Ensure session state lists are long enough
while len(st.session_state.scene_prompts) <= scene_index: st.session_state.scene_prompts.append("")
while len(st.session_state.generated_scene_assets) <= scene_index: st.session_state.generated_scene_assets.append(None)
st.session_state.scene_prompts[scene_index] = prompt_text_for_visual
# Filename base (extension will be added by visual_engine)
filename_base = f"scene_{scene_data.get('scene_number', scene_index+1)}_asset_v{version}"
runway_duration = scene_data.get('video_clip_duration_estimate_secs_감독', DEFAULT_SCENE_DURATION_SECS)
if runway_duration <= 0 : runway_duration = DEFAULT_SCENE_DURATION_SECS # Ensure positive duration
asset_result = st.session_state.visual_engine.generate_scene_asset(
image_prompt_text=prompt_text_for_visual, # This is generic, used for DALL-E or T2V
scene_data=scene_data,
scene_identifier_filename_base=filename_base,
generate_as_video_clip=generate_as_video,
runway_target_duration=runway_duration
# input_image_for_runway=None # TODO: Could be an enhancement
)
st.session_state.generated_scene_assets[scene_index] = asset_result # Store the whole dict
if asset_result and not asset_result['error'] and asset_result.get('path') and os.path.exists(asset_result['path']):
logger.info(f"Asset ({asset_result.get('type')}) generated for Scene {scene_data.get('scene_number', scene_index+1)}: {os.path.basename(asset_result['path'])}")
return True
else:
err_msg = asset_result.get('error_message', 'Unknown error') if asset_result else 'Asset result is None'
logger.warning(f"Asset generation FAILED for Scene {scene_data.get('scene_number', scene_index+1)}. Type attempted: {final_asset_type_decision}. Path was: {asset_result.get('path') if asset_result else 'N/A'}. Error: {err_msg}")
# Store a failure state
st.session_state.generated_scene_assets[scene_index] = {'path': None, 'type': 'none', 'error': True, 'error_message': err_msg, 'prompt_used': prompt_text_for_visual}
return False
# <<< MODIFIED END >>>
# --- UI Sidebar ---
with st.sidebar:
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()
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:
try:
status.write("Phase 1: Gemini crafting cinematic treatment... πŸ“œ"); logger.info("Phase 1: Cinematic Treatment Gen.")
treatment_prompt = create_cinematic_treatment_prompt(user_idea, genre, mood, num_scenes, actual_creative_guidance)
treatment_result_json_raw = st.session_state.gemini_handler.generate_story_breakdown(treatment_prompt) # Expect list of dicts
if not isinstance(treatment_result_json_raw, list) or not treatment_result_json_raw: raise ValueError("Gemini returned invalid scene list format.")
processed_scenes = []
for scene_data_from_gemini in treatment_result_json_raw:
scene_data_from_gemini['user_shot_type'] = scene_data_from_gemini.get('PROACTIVE_camera_work_감독', DEFAULT_SHOT_TYPE) # Default from Gemini's suggestion
scene_data_from_gemini['user_scene_duration_secs'] = scene_data_from_gemini.get('video_clip_duration_estimate_secs_감독', DEFAULT_SCENE_DURATION_SECS)
if scene_data_from_gemini['user_scene_duration_secs'] <=0: scene_data_from_gemini['user_scene_duration_secs'] = DEFAULT_SCENE_DURATION_SECS
scene_data_from_gemini['user_selected_asset_type'] = "Auto (Director's Choice)" # Default for UI
processed_scenes.append(scene_data_from_gemini)
st.session_state.story_treatment_scenes = processed_scenes
num_gen_scenes = len(st.session_state.story_treatment_scenes)
# <<< MODIFIED START >>>
st.session_state.scene_prompts = [""]*num_gen_scenes
st.session_state.generated_scene_assets = [None]*num_gen_scenes # Initialize list for asset dicts
# <<< MODIFIED END >>>
logger.info(f"Phase 1 complete. {num_gen_scenes} scenes."); status.update(label="Treatment complete! βœ… Generating visuals...", state="running")
status.write("Phase 2: Creating visual assets (Image/Video)... πŸ–ΌοΈπŸŽ¬"); logger.info("Phase 2: Visual Asset Gen.")
visual_successes = 0
for i, sc_data in enumerate(st.session_state.story_treatment_scenes):
sc_num_log = sc_data.get('scene_number', i+1)
status.write(f" Asset for Scene {sc_num_log}..."); logger.info(f" Processing asset for Scene {sc_num_log}.")
# <<< MODIFIED START >>> : Calling new function
if generate_asset_for_scene_core(i, sc_data, version=1): # Default to 'Auto' asset type for initial gen
visual_successes += 1
# <<< MODIFIED END >>>
current_status_label_ph2 = "Visual assets ready! "
next_step_state = "running"
if visual_successes == 0 and num_gen_scenes > 0:
logger.error("Visual asset gen failed for all scenes."); current_status_label_ph2 = "Asset gen FAILED for all scenes."; next_step_state="error";
status.update(label=current_status_label_ph2, state=next_step_state, expanded=True); st.stop()
elif visual_successes < num_gen_scenes:
logger.warning(f"Assets partially generated ({visual_successes}/{num_gen_scenes})."); current_status_label_ph2 = f"Assets partially generated ({visual_successes}/{num_gen_scenes}). "
status.update(label=f"{current_status_label_ph2}Generating narration script...", state=next_step_state)
if next_step_state == "error": st.stop()
status.write("Phase 3: Generating narration script..."); logger.info("Phase 3: Narration Script Gen.")
voice_style_for_prompt = st.session_state.get("selected_voice_style_for_generation", "cinematic_trailer")
narr_prompt = create_narration_script_prompt_enhanced(st.session_state.story_treatment_scenes, mood, genre, voice_style_for_prompt)
st.session_state.narration_script_display = st.session_state.gemini_handler.generate_image_prompt(narr_prompt) # This generates a string
logger.info("Narration script generated."); status.update(label="Narration script ready! Synthesizing voice...", state="running")
status.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_label = "All components ready! Storyboard below. πŸš€"
final_state_val = "complete"
if not st.session_state.overall_narration_audio_path:
final_label = f"{current_status_label_ph2}Storyboard ready (Voiceover skipped or failed)."
logger.warning("Voiceover was skipped or failed.")
else: logger.info("Voiceover generated successfully.")
status.update(label=final_label, state=final_state_val, expanded=False)
except ValueError as ve: logger.error(f"ValueError: {ve}", exc_info=True); status.update(label=f"Input or Gemini response error: {ve}", state="error", expanded=True);
except Exception as e: logger.error(f"Unhandled Exception: {e}", exc_info=True); status.update(label=f"An unexpected error occurred: {e}", state="error", expanded=True);
st.markdown("---"); st.markdown("### Fine-Tuning Options")
# ... (Character, Global Style, Voice expanders - no changes needed here for this fix) ...
with st.expander("Define Characters", expanded=False):
char_name = st.text_input("Character Name", key="char_name_adv_ultra_v5"); char_desc = st.text_area("Visual Description", key="char_desc_adv_ultra_v5", height=100, placeholder="e.g., Jax: rugged male astronaut...")
if st.button("Save Character", key="add_char_adv_ultra_v5"):
if char_name and char_desc: st.session_state.character_definitions[char_name.strip().lower()] = char_desc.strip(); st.success(f"Char '{char_name.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):
presets = { "Default (Director's Choice)": "", "Hyper-Realistic Gritty Noir": "hyper-realistic gritty neo-noir, extreme detail, deep dynamic shadows, complex reflections on wet surfaces, cinematic film grain, desaturated palette with isolated vibrant neon accents (e.g. red, cyan), anamorphic lens distortion, atmospheric haze.", "Surreal Dreamscape Fantasy": "surreal dreamscape, epic fantasy elements, painterly with photorealistic details, impossible architecture, bioluminescent flora, otherworldly color palette (e.g., magenta skies, turquoise rivers), style of Roger Dean meets ZdzisΕ‚aw BeksiΕ„ski.", "Vintage Analog Sci-Fi": "70s/80s analog sci-fi film aesthetic, tangible practical effects look, subtle light leaks, lens flares, warm filmic tones mixed with cool blues, detailed retro-futuristic technology with chunky buttons and CRT screens."}
sel_preset = st.selectbox("Base Style Preset:", options=list(presets.keys()), key="style_preset_adv_ultra_v5")
custom_kw = st.text_area("Additional Custom Style Keywords:", key="custom_style_adv_ultra_v5", height=80, placeholder="e.g., 'Dutch angle'")
cur_style = st.session_state.global_style_additions
if st.button("Apply Global Styles", key="apply_styles_adv_ultra_v5"):
final_s = presets[sel_preset];
if custom_kw.strip(): final_s = f"{final_s}, {custom_kw.strip()}" if final_s else custom_kw.strip()
st.session_state.global_style_additions = final_s.strip(); cur_style = final_s.strip()
if cur_style: st.success("Global styles applied!")
else: st.info("Global style additions cleared.")
if cur_style: st.caption(f"Active global styles: \"{cur_style}\"")
with st.expander("Voice & Narration Style", expanded=False):
default_voice_from_engine = "Rachel"
if hasattr(st.session_state, 'visual_engine') and st.session_state.visual_engine:
default_voice_from_engine = st.session_state.visual_engine.elevenlabs_voice_id
user_voice_id_override = st.text_input(
"ElevenLabs Voice ID (optional override):",
value=default_voice_from_engine,
key="el_voice_id_override_v5",
help=f"Defaulting to '{default_voice_from_engine}' from secrets/config. Enter a specific Voice ID from your ElevenLabs account to override."
)
prompt_v_styles = {"Cinematic Trailer": "cinematic_trailer", "Neutral Documentary": "documentary_neutral", "Character Introspection": "introspective_character"}
sel_prompt_v_style_key = st.selectbox("Narration Script Style:", list(prompt_v_styles.keys()), key="narr_style_sel_v5", index=0)
if st.button("Set Narrator Voice & Style", key="set_voice_btn_ultra_v5"):
final_voice_id_to_use = user_voice_id_override.strip()
if not final_voice_id_to_use:
final_voice_id_to_use = st.session_state.get("ELEVENLABS_VOICE_ID_CONFIG", "Rachel")
if hasattr(st.session_state, 'visual_engine'):
st.session_state.visual_engine.elevenlabs_voice_id = final_voice_id_to_use
st.session_state.selected_voice_style_for_generation = prompt_v_styles[sel_prompt_v_style_key]
st.success(f"Narrator Voice ID set to: {final_voice_id_to_use}. Script Style: {sel_prompt_v_style_key}")
logger.info(f"User updated ElevenLabs Voice ID to: {final_voice_id_to_use}, Script Style: {sel_prompt_v_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, scene_content_display in enumerate(st.session_state.story_treatment_scenes):
scene_n = scene_content_display.get('scene_number', i_main + 1); scene_t = scene_content_display.get('scene_title', 'Untitled')
key_base = f"s{scene_n}_{''.join(filter(str.isalnum, scene_t[:10]))}_v5_{i_main}"
if "director_note" in scene_content_display and scene_content_display['director_note']: st.info(f"🎬 Director Note S{scene_n}: {scene_content_display['director_note']}")
st.subheader(f"SCENE {scene_n}: {scene_t.upper()}"); col_d, col_v = st.columns([0.45, 0.55])
with col_d: # Treatment and Controls Column
with st.expander("πŸ“ Scene Treatment & Controls", expanded=True):
# Display scene textual details (emotional_beat, setting, etc.)
st.markdown(f"**Beat:** {scene_content_display.get('emotional_beat', 'N/A')}")
st.markdown(f"**Setting:** {scene_content_display.get('setting_description', 'N/A')}")
st.markdown(f"**Chars:** {', '.join(scene_content_display.get('characters_involved', ['N/A']))}")
st.markdown(f"**Focus Moment:** _{scene_content_display.get('character_focus_moment', 'N/A')}_")
st.markdown(f"**Plot Beat:** {scene_content_display.get('key_plot_beat', 'N/A')}")
st.markdown(f"**Dialogue Hook:** `\"{scene_content_display.get('suggested_dialogue_hook', '...')}\"`")
st.markdown("---")
st.markdown(f"**Dir. Visual Style:** _{scene_content_display.get('PROACTIVE_visual_style_감독', 'N/A')}_")
st.markdown(f"**Dir. Camera:** _{scene_content_display.get('PROACTIVE_camera_work_감독', 'N/A')}_")
st.markdown(f"**Dir. Sound:** _{scene_content_display.get('PROACTIVE_sound_design_감독', 'N/A')}_")
st.markdown("---")
st.markdown("##### Shot, Pacing & Asset Controls")
# User Shot Type (Camera Angle)
current_shot_type = st.session_state.story_treatment_scenes[i_main].get('user_shot_type', DEFAULT_SHOT_TYPE)
try: shot_type_index = SHOT_TYPES_OPTIONS.index(current_shot_type)
except ValueError: shot_type_index = SHOT_TYPES_OPTIONS.index(DEFAULT_SHOT_TYPE)
new_shot_type = st.selectbox("Dominant Shot Type:", options=SHOT_TYPES_OPTIONS, index=shot_type_index, key=f"shot_type_widget_{key_base}")
if new_shot_type != current_shot_type:
st.session_state.story_treatment_scenes[i_main]['user_shot_type'] = new_shot_type
# Consider if a re-run is needed or if DALL-E prompt should be updated based on this
# User Scene Duration
current_duration = st.session_state.story_treatment_scenes[i_main].get('user_scene_duration_secs', DEFAULT_SCENE_DURATION_SECS)
new_duration = st.number_input("Scene Duration (seconds):", min_value=1, max_value=300, value=current_duration, step=1, key=f"duration_widget_{key_base}")
if new_duration != current_duration:
st.session_state.story_treatment_scenes[i_main]['user_scene_duration_secs'] = new_duration
# <<< MODIFIED START >>> : User Asset Type Selection
current_user_asset_type = st.session_state.story_treatment_scenes[i_main].get('user_selected_asset_type', "Auto (Director's Choice)")
try: asset_type_idx = ASSET_TYPE_OPTIONS.index(current_user_asset_type)
except ValueError: asset_type_idx = 0 # Default to Auto
new_user_asset_type = st.selectbox("Asset Type Override:", ASSET_TYPE_OPTIONS, index=asset_type_idx, key=f"asset_type_sel_{key_base}",
help="Choose 'Image' or 'Video Clip'. 'Auto' uses Gemini's suggestion.")
if new_user_asset_type != current_user_asset_type:
st.session_state.story_treatment_scenes[i_main]['user_selected_asset_type'] = new_user_asset_type
# This change will be picked up by regeneration buttons
# <<< MODIFIED END >>>
st.markdown("---")
# Display generated prompt for the asset
current_prompt_for_asset = st.session_state.scene_prompts[i_main] if i_main < len(st.session_state.scene_prompts) else None
if current_prompt_for_asset:
with st.popover("πŸ‘οΈ View Asset Generation Prompt"):
st.markdown(f"**Prompt used for current asset:**"); st.code(current_prompt_for_asset, language='text')
pexels_q = scene_content_display.get('pexels_search_query_감독', None)
if pexels_q: st.caption(f"Pexels Fallback Query: `{pexels_q}`")
with col_v: # Visuals Column
# <<< MODIFIED START >>> : Display logic for different asset types
current_asset_data = st.session_state.generated_scene_assets[i_main] if i_main < len(st.session_state.generated_scene_assets) else None
if current_asset_data and not current_asset_data.get('error') and current_asset_data.get('path') and os.path.exists(current_asset_data['path']):
asset_path = current_asset_data['path']
asset_type = current_asset_data.get('type', 'image') # Default to image if type missing
if asset_type == 'image':
st.image(asset_path, caption=f"Scene {scene_n} ({asset_type}): {scene_t}")
elif asset_type == 'video':
try:
with open(asset_path, 'rb') as vf: video_bytes = vf.read()
st.video(video_bytes, format="video/mp4", start_time=0)
st.caption(f"Scene {scene_n} ({asset_type}): {scene_t}")
except Exception as e_vid:
st.error(f"Error displaying video {asset_path}: {e_vid}")
logger.error(f"Error displaying video {asset_path}: {e_vid}", exc_info=True)
else:
st.warning(f"Unknown asset type '{asset_type}' for Scene {scene_n}.")
else: # No asset, or error during generation
if st.session_state.story_treatment_scenes: # Check if treatment exists
error_msg = current_asset_data.get('error_message', 'Visual pending or failed.') if current_asset_data else 'Visual pending or failed.'
st.caption(error_msg)
# <<< MODIFIED END >>>
with st.popover(f"✏️ Edit Scene {scene_n} Treatment"):
fb_script = st.text_area("Changes to treatment:", key=f"treat_fb_{key_base}", height=150)
if st.button(f"πŸ”„ Update Scene {scene_n} Treatment", key=f"regen_treat_btn_{key_base}"):
if fb_script:
with st.status(f"Updating Scene {scene_n} Treatment & Asset...", expanded=True) as s_treat_regen:
# Preserve user's shot type, duration, and asset type choices
user_shot_type = st.session_state.story_treatment_scenes[i_main]['user_shot_type']
user_duration = st.session_state.story_treatment_scenes[i_main]['user_scene_duration_secs']
user_asset_type_choice = st.session_state.story_treatment_scenes[i_main]['user_selected_asset_type']
prompt_text = create_scene_regeneration_prompt(scene_content_display, fb_script, st.session_state.story_treatment_scenes)
try:
updated_sc_data_from_gemini = st.session_state.gemini_handler.regenerate_scene_script_details(prompt_text)
# Merge, but prioritize user's UI choices for duration/shot/asset type
updated_sc_data = {**updated_sc_data_from_gemini} # Start with Gemini's new script
updated_sc_data['user_shot_type'] = user_shot_type
updated_sc_data['user_scene_duration_secs'] = user_duration
updated_sc_data['user_selected_asset_type'] = user_asset_type_choice
# Gemini might re-suggest asset type/duration, but user's direct settings take precedence for next gen
# We can log if Gemini's suggestion differs from user's explicit choice.
if updated_sc_data.get('suggested_asset_type_감독') != user_asset_type_choice and user_asset_type_choice != "Auto (Director's Choice)":
logger.info(f"Scene {scene_n}: User asset choice '{user_asset_type_choice}' overrides Gemini suggestion '{updated_sc_data.get('suggested_asset_type_감독')}'.")
st.session_state.story_treatment_scenes[i_main] = updated_sc_data
s_treat_regen.update(label="Treatment updated! Regenerating asset...", state="running")
v_num = 1
if current_asset_data and current_asset_data.get('path') and os.path.exists(current_asset_data['path']):
try: b,_=os.path.splitext(os.path.basename(current_asset_data['path'])); v_num = int(b.split('_v')[-1])+1 if '_v' in b else 2
except: v_num = 2
else: v_num = 1
# <<< MODIFIED START >>> : Call new function, pass user_selected_asset_type
if generate_asset_for_scene_core(i_main, updated_sc_data, version=v_num, user_selected_asset_type=user_asset_type_choice):
s_treat_regen.update(label="Treatment & Asset Updated! πŸŽ‰", state="complete", expanded=False)
else: s_treat_regen.update(label="Treatment updated, asset failed.", state="complete", expanded=False)
# <<< MODIFIED END >>>
st.rerun()
except Exception as e_regen: s_treat_regen.update(label=f"Error: {e_regen}", state="error"); logger.error(f"Scene treatment regen error: {e_regen}", exc_info=True)
else: st.warning("Please provide feedback.")
with st.popover(f"🎨 Edit Scene {scene_n} Visual Prompt"):
prompt_to_edit = st.session_state.scene_prompts[i_main] if i_main < len(st.session_state.scene_prompts) else "No prompt generated yet."
st.caption("Current Asset Generation Prompt:"); st.code(prompt_to_edit, language='text')
fb_visual = st.text_area("Changes for asset generation prompt:", key=f"visual_fb_{key_base}", height=150)
if st.button(f"πŸ”„ Update Scene {scene_n} Asset", key=f"regen_visual_btn_{key_base}"):
if fb_visual:
with st.status(f"Refining prompt & asset for Scene {scene_n}...", expanded=True) as s_visual_regen:
user_asset_type_choice = st.session_state.story_treatment_scenes[i_main]['user_selected_asset_type']
is_video_prompt = (user_asset_type_choice == "Video Clip") or \
(user_asset_type_choice == "Auto (Director's Choice)" and \
scene_content_display.get('suggested_asset_type_감독') == 'video_clip')
# Note: Visual regeneration prompt is primarily for DALL-E (images).
# For video, we might need a different refinement strategy or just regenerate with the same prompt construction.
# For simplicity here, if it's a video, we'll regenerate the prompt using standard construction.
# If it's an image, we use Gemini to refine the DALL-E prompt.
new_asset_gen_prompt = ""
if not is_video_prompt : # Refining an image prompt
ref_req_prompt_for_gemini = create_visual_regeneration_prompt(prompt_to_edit, fb_visual, scene_content_display,
st.session_state.character_definitions, st.session_state.global_style_additions)
try:
new_asset_gen_prompt = st.session_state.gemini_handler.refine_image_prompt_from_feedback(ref_req_prompt_for_gemini)
st.session_state.scene_prompts[i_main] = new_asset_gen_prompt
s_visual_regen.update(label="Image prompt refined by Gemini! Regenerating asset...", state="running")
except Exception as e_gemini_refine:
s_visual_regen.update(label=f"Error refining prompt: {e_gemini_refine}", state="error");
logger.error(f"Visual prompt refinement error: {e_gemini_refine}", exc_info=True)
continue # Skip asset generation if prompt refinement failed
else: # For video, or auto choosing video, reconstruct the prompt
new_asset_gen_prompt = construct_text_to_video_prompt(scene_content_display, st.session_state.character_definitions, st.session_state.global_style_additions)
st.session_state.scene_prompts[i_main] = new_asset_gen_prompt
s_visual_regen.update(label="Video prompt reconstructed! Regenerating asset...", state="running")
v_num = 1
if current_asset_data and current_asset_data.get('path') and os.path.exists(current_asset_data['path']):
try: b,_=os.path.splitext(os.path.basename(current_asset_data['path'])); v_num = int(b.split('_v')[-1])+1 if '_v' in b else 2
except: v_num=2
else: v_num = 1
# <<< MODIFIED START >>> : Call new function
# Pass the current scene_content_display as its prompt might have changed.
# User asset type choice from the scene data for consistency
if generate_asset_for_scene_core(i_main, st.session_state.story_treatment_scenes[i_main], version=v_num, user_selected_asset_type=user_asset_type_choice):
s_visual_regen.update(label="Asset Updated! πŸŽ‰", state="complete", expanded=False)
else: s_visual_regen.update(label="Prompt updated, asset regeneration failed.", state="complete", expanded=False)
# <<< MODIFIED END >>>
st.rerun()
else: st.warning("Please provide feedback.")
st.markdown("---")
# Video Assembly Button
# <<< MODIFIED START >>> : Check generated_scene_assets and use its data
if st.session_state.story_treatment_scenes and any(asset_info and not asset_info.get('error') and asset_info.get('path') for asset_info in st.session_state.generated_scene_assets if asset_info is not None):
if st.button("🎬 Assemble Narrated Cinematic Animatic", key="assemble_ultra_video_btn_v5", type="primary", use_container_width=True):
with st.status("Assembling Ultra Animatic...", expanded=True) as status_vid:
assets_for_video_assembly = []
for i_v, sc_c in enumerate(st.session_state.story_treatment_scenes):
asset_info = st.session_state.generated_scene_assets[i_v] if i_v < len(st.session_state.generated_scene_assets) else None
if asset_info and not asset_info.get('error') and asset_info.get('path') and os.path.exists(asset_info['path']):
assets_for_video_assembly.append({
'path': asset_info['path'],
'type': asset_info.get('type', 'image'), # Default to image if type missing
'scene_num': sc_c.get('scene_number', i_v + 1),
'key_action': sc_c.get('key_plot_beat', ''),
'duration': sc_c.get('user_scene_duration_secs', DEFAULT_SCENE_DURATION_SECS) # Use user-set duration
})
status_vid.write(f"Adding Scene {sc_c.get('scene_number', i_v + 1)} ({asset_info.get('type')}).")
else:
logger.warning(f"Skipping Scene {sc_c.get('scene_number', i_v+1)} for video: No valid asset.")
if assets_for_video_assembly:
status_vid.write("Calling video engine...");
st.session_state.video_path = st.session_state.visual_engine.assemble_animatic_from_assets( # Changed method name
asset_data_list=assets_for_video_assembly, # Pass the list of asset dicts
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_vid.update(label="Ultra animatic assembled! πŸŽ‰", state="complete", expanded=False); st.balloons()
else:
status_vid.update(label="Video assembly failed. Check logs.", state="error", expanded=False); logger.error("Video assembly returned None or file does not exist.")
else:
status_vid.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 video.")
# <<< MODIFIED END >>>
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: video_bytes = vf_obj.read()
st.video(video_bytes, format="video/mp4")
st.download_button(label="Download Ultra Animatic", data=video_bytes, file_name=os.path.basename(st.session_state.video_path), mime="video/mp4", use_container_width=True, key="download_ultra_video_btn_v5" )
except Exception as e: st.error(f"Error displaying video: {e}"); logger.error(f"Error displaying video: {e}", exc_info=True)
# --- Footer ---
st.sidebar.markdown("---"); st.sidebar.caption("CineGen AI Ultra+ | Visionary Cinematic Pre-Production")