CingenAI / app.py
mgbam's picture
Update app.py
2ffd5d9 verified
raw
history blame
21.1 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_story_breakdown_prompt,
create_image_prompt_from_scene_data,
create_scene_regeneration_prompt,
create_visual_regeneration_prompt
)
import os
# --- Configuration & Initialization ---
st.set_page_config(page_title="CineGen AI Pro", layout="wide", initial_sidebar_state="expanded")
# --- Global State Variables ---
if 'GEMINI_API_KEY' not in st.session_state:
try: st.session_state.GEMINI_API_KEY = st.secrets["GEMINI_API_KEY"]
except:
if "GEMINI_API_KEY" in os.environ: st.session_state.GEMINI_API_KEY = os.environ["GEMINI_API_KEY"]
else: st.error("GEMINI_API_KEY not found."); st.stop()
if 'gemini_handler' not in st.session_state:
st.session_state.gemini_handler = GeminiHandler(api_key=st.session_state.GEMINI_API_KEY)
if 'visual_engine' not in st.session_state:
st.session_state.visual_engine = VisualEngine(output_dir="temp_cinegen_media")
openai_key = None
try:
if "OPENAI_API_KEY" in st.secrets: openai_key = st.secrets["OPENAI_API_KEY"]
except AttributeError: print("st.secrets not available. Checking env for OpenAI key.")
except Exception as e: print(f"Error accessing st.secrets for OPENAI_API_KEY: {e}")
if not openai_key and "OPENAI_API_KEY" in os.environ: openai_key = os.environ["OPENAI_API_KEY"]
st.session_state.visual_engine.set_openai_api_key(openai_key)
if not openai_key and 'openai_warning_shown' not in st.session_state:
st.sidebar.warning("OpenAI Key missing. DALL-E disabled, using placeholders.")
st.session_state.openai_warning_shown = True # Show only once per session
elif openai_key and 'openai_success_shown' not in st.session_state:
# st.sidebar.caption("OpenAI Key loaded. DALL-E ready.") # Can be a bit noisy
st.session_state.openai_success_shown = True
if 'story_scenes' not in st.session_state: st.session_state.story_scenes = []
if 'scene_image_prompts' not in st.session_state: st.session_state.scene_image_prompts = []
if 'generated_images_paths' not in st.session_state: st.session_state.generated_images_paths = []
if 'video_path' not in st.session_state: st.session_state.video_path = None
if 'character_definitions' not in st.session_state: st.session_state.character_definitions = {}
if 'style_reference_description' not in st.session_state: st.session_state.style_reference_description = ""
# --- Helper Functions ---
def initialize_new_story(): # Remains the same
st.session_state.story_scenes = []
st.session_state.scene_image_prompts = []
st.session_state.generated_images_paths = []
st.session_state.video_path = None
def generate_visual_for_scene_wrapper(scene_index, scene_data, is_regeneration=False, version_count=1):
scene_num_for_log = scene_data.get('scene_number', scene_index + 1)
textual_image_prompt = ""
if is_regeneration and scene_index < len(st.session_state.scene_image_prompts) and st.session_state.scene_image_prompts[scene_index]:
textual_image_prompt = st.session_state.scene_image_prompts[scene_index] # Use the already refined prompt
else: # Initial generation
textual_image_prompt = create_image_prompt_from_scene_data(
scene_data,
st.session_state.character_definitions,
st.session_state.style_reference_description
)
if not textual_image_prompt: return False
# Update session state with the textual prompt used (might be initial or refined)
if scene_index >= len(st.session_state.scene_image_prompts):
while len(st.session_state.scene_image_prompts) <= scene_index: st.session_state.scene_image_prompts.append("")
st.session_state.scene_image_prompts[scene_index] = textual_image_prompt
image_filename = f"scene_{scene_num_for_log}_visual_v{version_count}.png"
generated_image_path = st.session_state.visual_engine.generate_image_visual(textual_image_prompt, image_filename)
while len(st.session_state.generated_images_paths) <= scene_index: st.session_state.generated_images_paths.append(None)
if generated_image_path and os.path.exists(generated_image_path):
st.session_state.generated_images_paths[scene_index] = generated_image_path
return True
else:
st.session_state.generated_images_paths[scene_index] = None
return False
# --- UI Sidebar ---
with st.sidebar:
st.title("🎬 CineGen AI Pro")
st.markdown("### Creative Controls")
user_idea = st.text_area("Enter your core story idea:", "A lone astronaut discovers a glowing alien artifact on Mars, a sense of wonder and slight dread.", height=100, key="user_idea_input")
genre = st.selectbox("Genre:", ["Cyberpunk", "Sci-Fi", "Fantasy", "Noir", "Thriller", "Drama", "Horror"], index=0, key="genre_select")
mood = st.selectbox("Mood:", ["Wonder", "Suspenseful", "Mysterious", "Gritty", "Epic", "Dark", "Hopeful"], index=0, key="mood_select")
num_scenes_val = st.slider("Number of Scenes:", 1, 3, 1, key="num_scenes_slider") # Max 3 for DALL-E cost/time
if st.button("✨ Generate Full Story Concept", type="primary", key="generate_full_story_btn", use_container_width=True):
initialize_new_story()
if not user_idea.strip(): st.warning("Please enter a story idea.")
else:
with st.status("Generating story and visuals...", expanded=True) as status_main:
st.write("Phase 1: Drafting script with Gemini... πŸ“œ")
story_prompt_text = create_story_breakdown_prompt(user_idea, genre, mood, num_scenes_val)
try:
st.session_state.story_scenes = st.session_state.gemini_handler.generate_story_breakdown(story_prompt_text)
status_main.update(label="Script complete! βœ… Generating visuals...", state="running")
num_actual_scenes = len(st.session_state.story_scenes)
st.session_state.scene_image_prompts = [""] * num_actual_scenes
st.session_state.generated_images_paths = [None] * num_actual_scenes
except Exception as e:
status_main.update(label=f"Script generation failed: {e}", state="error"); st.stop()
if st.session_state.story_scenes:
st.write("Phase 2: Creating visuals with DALL-E (this can take time)... πŸ–ΌοΈ")
success_count = 0
for i, scene_data in enumerate(st.session_state.story_scenes):
st.write(f" Processing Scene {scene_data.get('scene_number', i + 1)}...")
if generate_visual_for_scene_wrapper(i, scene_data, version_count=1): success_count +=1
if success_count == len(st.session_state.story_scenes): status_main.update(label="All done! Story and visuals ready. πŸŽ‰", state="complete", expanded=False)
elif success_count > 0: status_main.update(label=f"{success_count}/{len(st.session_state.story_scenes)} visuals OK.", state="warning", expanded=False)
else: status_main.update(label="Visual generation failed for all scenes.", state="error", expanded=False)
st.markdown("---")
st.markdown("### Advanced Options")
with st.expander("Character Consistency", expanded=False):
char_name_input = st.text_input("Character Name (e.g., Jax)", key="char_name_adv_input")
char_desc_input = st.text_area("Character Description (e.g., 'male astronaut, rugged, dark short hair, blue eyes, wearing a worn white and orange spacesuit with a cracked visor helmet slung at his hip')", key="char_desc_adv_input", height=120)
if st.button("Add/Update Character", key="add_char_adv_btn"):
if char_name_input and char_desc_input:
st.session_state.character_definitions[char_name_input.strip().lower()] = char_desc_input.strip() # Use lower for key
st.success(f"Character '{char_name_input.strip()}' defined.")
else: st.warning("Provide both name and description.")
if st.session_state.character_definitions:
st.caption("Defined Characters:")
for char_key, desc_val in st.session_state.character_definitions.items(): st.markdown(f"**{char_key.title()}:** _{desc_val}_")
with st.expander("Style Controls", expanded=True): # Expanded by default
predefined_styles = {
"Default (Cinematic Photorealism)": "",
"Gritty Neo-Noir": "neo-noir aesthetic, gritty realism, deep shadows, rain-slicked streets, desaturated colors with vibrant neon highlights, anamorphic lens flares, film grain",
"Epic Fantasy Matte Painting": "epic fantasy matte painting style, vast landscapes, dramatic skies, painterly textures, rich colors, Lord of the Rings inspired, highly detailed",
"Impressionistic Dream": "impressionistic oil painting style, visible brushstrokes, soft focus, dreamlike atmosphere, pastel color palette, ethereal lighting",
"Vintage Sci-Fi (70s/80s Film)": "retro sci-fi aesthetic, analog film look, subtle film grain, practical effects feel, slightly desaturated colors, chrome and plastic textures, classic 70s/80s anamorphic lens flares, detailed spaceship interiors or alien landscapes",
"Modern Anime Key Visual": "high-detail modern anime key visual style, dynamic compositions, vibrant saturated colors, crisp cel-shaded look with intricate lighting and deep shadows, expressive character designs with detailed eyes"
}
selected_preset = st.selectbox("Base Style Preset:", options=list(predefined_styles.keys()), key="style_preset_select_adv")
custom_keywords = st.text_area("Add Custom Style Keywords:", key="custom_style_keywords_adv_input", height=80, placeholder="e.g., 'Dutch angle, fisheye lens, sepia tone'")
current_style_desc = st.session_state.style_reference_description
if st.button("Apply & Set Styles", key="apply_styles_adv_btn"):
final_desc = predefined_styles[selected_preset]
if custom_keywords.strip():
final_desc = f"{final_desc}, {custom_keywords.strip()}" if final_desc else custom_keywords.strip()
st.session_state.style_reference_description = final_desc.strip()
current_style_desc = final_desc.strip() # Update for immediate display
if current_style_desc: st.success("Styles applied! Regenerate story or visuals.")
else: st.info("Default style (no specific additions).")
if current_style_desc: st.caption(f"Active style prompt additions: \"{current_style_desc}\"")
else: st.caption("No specific style prompt additions active.")
# --- Main Content Area (largely same as previous full version, with minor tweaks) ---
st.header("πŸ“ Cinematic Storyboard")
if not st.session_state.story_scenes:
st.info("Enter your idea in the sidebar and click '✨ Generate Full Story Concept' to begin.")
else:
for i, scene_data_display in enumerate(st.session_state.story_scenes):
# ... (scene_num_display, unique_key_part, subheader, columns - same) ...
scene_num_display = scene_data_display.get('scene_number', i + 1)
action_summary = scene_data_display.get('key_action', f"scene{i}")
key_part_raw = ''.join(e for e in action_summary if e.isalnum() or e.isspace())
unique_key_part = key_part_raw.replace(" ", "_")[:20]
st.subheader(f"Scene {scene_num_display}: {scene_data_display.get('emotional_beat', 'Untitled Scene')}")
col1, col2 = st.columns([2, 3])
with col1: # Scene Details
with st.expander("Scene Details", expanded=True):
st.markdown(f"**Setting:** {scene_data_display.get('setting_description', 'N/A')}")
st.markdown(f"**Characters:** {', '.join(scene_data_display.get('characters_involved', []))}")
st.markdown(f"**Key Action:** {scene_data_display.get('key_action', 'N/A')}")
st.markdown(f"**Dialogue Snippet:** `\"{scene_data_display.get('dialogue_snippet', '...')}\"`")
st.markdown(f"**Visual Style Suggestion (from Gemini):** {scene_data_display.get('visual_style_suggestion', 'N/A')}")
st.markdown(f"**Camera Suggestion (from Gemini):** {scene_data_display.get('camera_angle_suggestion', 'N/A')}")
current_textual_prompt = st.session_state.scene_image_prompts[i] if i < len(st.session_state.scene_image_prompts) else None
if current_textual_prompt:
with st.popover("View Full Image Prompt"):
st.markdown(f"**Full Textual Prompt Sent to DALL-E:**"); st.code(current_textual_prompt, language='text')
with col2: # Image and Edit Buttons
current_image_path = st.session_state.generated_images_paths[i] if i < len(st.session_state.generated_images_paths) else None
if current_image_path and os.path.exists(current_image_path):
st.image(current_image_path, caption=f"Visual Concept for Scene {scene_num_display}")
else:
if st.session_state.story_scenes: st.caption("Visual for this scene is pending or failed.")
# Edit Popovers (logic remains largely the same, calls generate_visual_for_scene_wrapper)
with st.popover(f"✏️ Edit Scene {scene_num_display} Script"):
feedback_script = st.text_area("Describe script changes:", key=f"script_feedback_{unique_key_part}_{i}", height=100)
if st.button(f"πŸ”„ Update Scene {scene_num_display} Script", key=f"regen_script_btn_{unique_key_part}_{i}"):
if feedback_script:
with st.status(f"Updating Scene {scene_num_display}...", expanded=True) as status_script_regen:
regen_prompt = create_scene_regeneration_prompt(scene_data_display, feedback_script, st.session_state.story_scenes)
try:
updated_scene_data = st.session_state.gemini_handler.regenerate_scene_script_details(regen_prompt)
st.session_state.story_scenes[i] = updated_scene_data
status_script_regen.update(label="Script updated! Regenerating visual...", state="running")
version = 1
if current_image_path:
try: base, _ = os.path.splitext(os.path.basename(current_image_path)); version = int(base.split('_v')[-1]) + 1 if '_v' in base else 2
except: version = 2 # Default to v2 if parsing fails
if generate_visual_for_scene_wrapper(i, updated_scene_data, is_regeneration=False, version_count=version): # is_regeneration=False to use new scene_data for prompt
status_script_regen.update(label="Scene script & visual updated! πŸŽ‰", state="complete", expanded=False)
else: status_script_regen.update(label="Script updated, visual failed.", state="warning", expanded=False)
st.rerun()
except Exception as e: status_script_regen.update(label=f"Error: {e}", state="error")
else: st.warning("Please provide feedback.")
with st.popover(f"🎨 Edit Scene {scene_num_display} Visuals"):
prompt_for_edit_visuals = st.session_state.scene_image_prompts[i] if i < len(st.session_state.scene_image_prompts) else "No prompt."
st.caption("Current Full Image Prompt:"); st.code(prompt_for_edit_visuals, language='text')
feedback_visual = st.text_area("Describe visual changes for the prompt:", key=f"visual_feedback_{unique_key_part}_{i}", height=100)
if st.button(f"πŸ”„ Update Scene {scene_num_display} Visuals", key=f"regen_visual_btn_{unique_key_part}_{i}"):
if feedback_visual:
with st.status(f"Refining prompt & regenerating visual...", expanded=True) as status_visual_regen:
prompt_refinement_request = create_visual_regeneration_prompt(
prompt_for_edit_visuals, feedback_visual, scene_data_display,
st.session_state.character_definitions, st.session_state.style_reference_description
)
try:
refined_textual_image_prompt = st.session_state.gemini_handler.regenerate_image_prompt_from_feedback(prompt_refinement_request)
st.session_state.scene_image_prompts[i] = refined_textual_image_prompt # Update the prompt in session state
status_visual_regen.update(label="Prompt refined! Regenerating visual...", state="running")
version = 1
if current_image_path:
try: base, _ = os.path.splitext(os.path.basename(current_image_path)); version = int(base.split('_v')[-1]) + 1 if '_v' in base else 2
except: version = 2
if generate_visual_for_scene_wrapper(i, scene_data_display, is_regeneration=True, version_count=version): # is_regeneration=True to use the refined prompt
status_visual_regen.update(label="Visual updated! πŸŽ‰", state="complete", expanded=False)
else: status_visual_regen.update(label="Prompt refined, visual failed.", state="warning", expanded=False)
st.rerun()
except Exception as e: status_visual_regen.update(label=f"Error: {e}", state="error")
else: st.warning("Please provide feedback.")
st.markdown("---")
# Video Generation Button and Display (remains same as previous full app.py)
if st.session_state.story_scenes and any(p for p in st.session_state.generated_images_paths if p is not None):
if st.button("🎬 Assemble Enhanced Animatic", key="assemble_enhanced_video_btn", type="primary", use_container_width=True):
# ... (video assembly logic - same as before) ...
with st.status("Assembling enhanced animatic video...", expanded=True) as status_video:
image_data_for_video = []
for idx, scene_info in enumerate(st.session_state.story_scenes):
img_path = st.session_state.generated_images_paths[idx] if idx < len(st.session_state.generated_images_paths) else None
if img_path and os.path.exists(img_path):
image_data_for_video.append({
'path': img_path, 'scene_num': scene_info.get('scene_number', idx + 1),
'key_action': scene_info.get('key_action', '')
}); st.write(f"Adding Scene {scene_info.get('scene_number', idx + 1)} to video.")
if image_data_for_video:
st.write("Calling video engine..."); st.session_state.video_path = st.session_state.visual_engine.create_video_from_images(
image_data_for_video, output_filename="cinegen_pro_animatic_enhanced.mp4", duration_per_image=4, fps=24 )
if st.session_state.video_path and os.path.exists(st.session_state.video_path):
status_video.update(label="Enhanced animatic assembled! πŸŽ‰", state="complete", expanded=False); st.balloons()
else: status_video.update(label="Video assembly failed. Check logs.", state="error", expanded=False)
else: status_video.update(label="No valid images to assemble video.", state="error", expanded=False)
elif st.session_state.story_scenes: st.info("Generate visuals for scenes before assembling the video.")
if st.session_state.video_path and os.path.exists(st.session_state.video_path):
st.header("🎬 Generated Animatic Video")
try:
with open(st.session_state.video_path, 'rb') as video_file_obj: video_bytes_content = video_file_obj.read()
st.video(video_bytes_content, format="video/mp4")
with open(st.session_state.video_path, "rb") as fp_download_video:
st.download_button(label="Download Enhanced Animatic", data=fp_download_video,
file_name=os.path.basename(st.session_state.video_path), mime="video/mp4",
use_container_width=True, key="download_video_btn_enhanced" )
except Exception as e: st.error(f"Error displaying video: {e}")
# --- Footer ---
st.sidebar.markdown("---")
st.sidebar.caption("CineGen AI Pro | Powered by Gemini & Streamlit.")