Hackathon / app.py
SameerArz's picture
Create app.py
c0d2d56 verified
raw
history blame
10.6 kB
#!/usr/bin/env python3
import streamlit as st
from gradio_client import Client
from PIL import Image
import moviepy.editor as mp
from natsort import natsorted
from pydantic import BaseModel, Field
from typing import List, Dict, Type, Optional, TypedDict
from langgraph.graph import StateGraph, START, END
from langchain_groq import ChatGroq
from langchain_core.messages import SystemMessage
import os
from dotenv import load_dotenv
# Load environment variables
load_dotenv()
# Constants
HF_TOKEN = os.getenv("HF_TOKEN")
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
IMAGE_GENERATION_SPACE_NAME = "habib926653/stabilityai-stable-diffusion-3.5-large-turbo"
SUPPORTED_FORMATS = ["mp3", "wav", "ogg", "flac", "aac", "m4a"]
# Pydantic Models
class SingleScene(BaseModel):
text: str = Field(description="Actual Segment of text(a scene) from the complete story")
image_prompts: List[str] = Field(
description="""List of detailed and descriptive image prompts for the segment
prompt format: [theme: {atmosphere/mood}] [style: {artistic/photorealistic}] [focus: {main subject}] [details: {specific elements}] [lighting: {day/night/mystic}] [perspective: {close-up/wide-angle}]"
Example: "theme: eerie forest | style: cinematic realism | focus: abandoned cabin | details: broken windows, overgrown vines | lighting: moonlit fog | perspective: wide-angle shot"
"""
)
class ScenesResponseSchema(BaseModel):
scenes: List[SingleScene]
# Structured Output Extractor
class State(TypedDict):
messages: list
output: Optional[BaseModel]
class StructuredOutputExtractor:
def __init__(self, response_schema: Type[BaseModel]):
self.response_schema = response_schema
self.llm = ChatGroq(model="deepseek-r1-distill-llama-70b", api_key=GROQ_API_KEY)
self.structured_llm = self.llm.with_structured_output(response_schema)
self._build_graph()
def _build_graph(self):
graph_builder = StateGraph(State)
graph_builder.add_node("extract", self._extract_structured_info)
graph_builder.add_edge(START, "extract")
graph_builder.add_edge("extract", END)
self.graph = graph_builder.compile()
def _extract_structured_info(self, state: dict):
query = state['messages'][-1].content
try:
output = self.structured_llm.invoke(query)
return {"output": output}
except Exception as e:
st.error(f"Error during extraction: {e}")
return {"output": None}
def extract(self, query: str) -> Optional[BaseModel]:
result = self.graph.invoke({"messages": [SystemMessage(content=query)]})
return result.get('output')
# Utility Functions
def calculate_read_time(text: str, words_per_minute: int = 155) -> str:
try:
if not text or not isinstance(text, str):
return "Invalid input: Text must be a non-empty string."
words = text.split()
word_count = len(words)
total_seconds = (word_count / words_per_minute) * 60
hours = int(total_seconds // 3600)
minutes = int((total_seconds % 3600) // 60)
seconds = int(total_seconds % 60)
if hours > 0:
return f"Reading time: {hours} hour(s), {minutes} minute(s), and {seconds} second(s)."
elif minutes > 0:
return f"Reading time: {minutes} minute(s) and {seconds} second(s)."
else:
return f"Reading time: {seconds} second(s)."
except Exception as e:
return f"An error occurred: {e}"
def get_scenes(text_script: str):
read_time = calculate_read_time(text_script)
prompt = f"""
ROLE: Story to Scene Generator
Tasks: For the given story
1. Read it Completely and Understand the Complete Context
2. Rewrite the story in tiny scenes(but without even changing a word) with highly detailed and context aware list of image prompts to visualize each scene
3. Never Describe complete scene in a single image prompt use multiple prompts
RULE OF THUMB: 12 image prompts / 1 min audio
Estimated Read Time: {read_time}\n\n
Complete Story: {text_script}
"""
extractor = StructuredOutputExtractor(response_schema=ScenesResponseSchema)
result = extractor.extract(prompt)
return result.model_dump() if result else {}
def generate_audio(text, language_code, speaker, path='test_audio.mp3'):
try:
client = Client("habib926653/Multilingual-TTS")
result = client.predict(
text=text,
language_code=language_code,
speaker=speaker,
api_name="/text_to_speech_edge"
)
audio_file_path = result[1]
with open(audio_file_path, 'rb') as f:
audio_bytes = f.read()
with open(path, 'wb') as f:
f.write(audio_bytes)
return {"audio_file": path}
except Exception as e:
st.error(f"Error during audio generation: {e}")
return {"error": str(e)}
def generate_image(prompt, path='test_image.png'):
try:
client = Client(IMAGE_GENERATION_SPACE_NAME, hf_token=HF_TOKEN)
result = client.predict(
prompt=prompt,
width=1280,
height=720,
api_name="/generate_image"
)
image = Image.open(result)
image.save(path)
return result
except Exception as e:
st.error(f"Error during image generation: {e}")
return {"error": str(e)}
def generate_video_assets(scenes: Dict, language: str, speaker: str, base_path: str = "media") -> str:
try:
if not os.path.exists(base_path):
os.makedirs(base_path)
scenes_list = scenes.get("scenes", [])
video_folder = os.path.join(base_path, f"video_{len(os.listdir(base_path)) + 1}")
os.makedirs(video_folder, exist_ok=True)
images_folder = os.path.join(video_folder, "images")
audio_folder = os.path.join(video_folder, "audio")
os.makedirs(images_folder, exist_ok=True)
os.makedirs(audio_folder, exist_ok=True)
for scene_count, scene in enumerate(scenes_list):
text = scene.get("text", "")
image_prompts = scene.get("image_prompts", [])
audio_path = os.path.join(audio_folder, f"scene_{scene_count + 1}.mp3")
audio_result = generate_audio(text, language, speaker, path=audio_path)
if "error" in audio_result:
continue
scene_images_folder = os.path.join(images_folder, f"scene_{scene_count + 1}")
os.makedirs(scene_images_folder, exist_ok=True)
for count, prompt in enumerate(image_prompts):
image_path = os.path.join(scene_images_folder, f"scene_{scene_count + 1}_image_{count + 1}.png")
generate_image(prompt=prompt, path=image_path)
return video_folder
except Exception as e:
st.error(f"Error during video asset generation: {e}")
return ""
def generate_video(video_folder: str, output_filename: str = "final_video.mp4"):
try:
audio_folder = os.path.join(video_folder, "audio")
images_folder = os.path.join(video_folder, "images")
final_clips = []
scene_folders = [
os.path.join(images_folder, scene)
for scene in natsorted(os.listdir(images_folder))
if os.path.isdir(os.path.join(images_folder, scene))
]
for scene_path in scene_folders:
scene_name = os.path.basename(scene_path)
audio_path = os.path.join(audio_folder, f"{scene_name}.mp3")
if not os.path.exists(audio_path):
continue
image_files = natsorted([
os.path.join(scene_path, img)
for img in os.listdir(scene_path)
if img.lower().endswith(('.png', '.jpg', '.jpeg'))
])
if not image_files:
continue
audio_clip = mp.AudioFileClip(audio_path)
duration_per_image = audio_clip.duration / len(image_files)
image_clips = [mp.ImageClip(img).set_duration(duration_per_image) for img in image_files]
scene_video = mp.concatenate_videoclips(image_clips, method="compose").set_audio(audio_clip)
final_clips.append(scene_video)
if not final_clips:
st.error("No valid scenes processed.")
return None
final_video = mp.concatenate_videoclips(final_clips, method="compose")
output_path = os.path.join(video_folder, output_filename)
final_video.write_videofile(output_path, fps=24, codec='libx264')
return output_path
except Exception as e:
st.error(f"Error during video generation: {e}")
return None
# Streamlit App
def main():
st.markdown("<h1 style='text-align: center;'>Text to Video Generator</h1>", unsafe_allow_html=True)
st.markdown("<p style='text-align: center;'>Leave a Like if it works for you! ❤️</p>", unsafe_allow_html=True)
text_script = st.text_area("Enter your script/story (max 1500 characters):", max_chars=1500)
language = st.selectbox("Choose Language:", ["Urdu", "English"])
client = Client("habib926653/Multilingual-TTS")
speakers_response = client.predict(language=language, api_name="/get_speakers")
speakers = [choice[0] for choice in speakers_response["choices"]]
selected_speaker = st.selectbox("Choose Speaker:", speakers)
if st.button("Generate Video"):
if text_script:
with st.spinner("Generating video... This may take a few minutes."):
scenes = get_scenes(text_script)
if not scenes:
st.error("Failed to generate scenes.")
else:
video_assets_folder = generate_video_assets(scenes, language, selected_speaker)
if video_assets_folder:
generated_video_path = generate_video(video_assets_folder)
if generated_video_path:
st.success("Video generated successfully!")
st.video(generated_video_path)
else:
st.warning("Please enter some text to generate a video.")
st.markdown("### 🔥 See How It Works (Example)")
example_script = """
One hot summer day, a thirsty crow was flying in search of water. He looked everywhere, but he couldn't find a single drop. Tired and exhausted, he finally spotted a clay pot with a little water at the bottom.
"""
st.markdown(f"**Example Script:** {example_script}")
if __name__ == "__main__":
main()