app.py

import gradio as gr
import torch
import torchaudio
from diffusers import StableDiffusionPipeline
from TTS.api import TTS
import moviepy.editor as mp
import numpy as np
import os
from PIL import Image

def estimate_chunk_durations(text, words_per_second=2.5, min_sec=5, max_sec=7):
    words = text.split()
    chunks = []
    current_chunk = []
    current_duration = 0
    for word in words:
        current_chunk.append(word)
        current_duration += 1 / words_per_second
        if current_duration >= min_sec:
            if current_duration >= max_sec or len(current_chunk) > 20:
                chunks.append(" ".join(current_chunk))
                current_chunk = []
                current_duration = 0
    if current_chunk:
        chunks.append(" ".join(current_chunk))
    return chunks

def generate_speech(text):
    #tts = TTS("tts_models/en/ljspeech/tacotron2-DDC")
    tts = TTS("tts_models/en/ljspeech/glow-tts")

    #tts.synthesizer.model.decoder.max_decoder_steps = 30000  # Increase limit
    wav_path = "speech.wav"
    tts.tts_to_file(text=text, file_path=wav_path)
    return wav_path

def generate_images(chunks, image_size=(640, 480), use_diffusion=True, num_steps=40):
    image_paths = []
    if use_diffusion:
        pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4")
        pipe.to("cuda" if torch.cuda.is_available() else "cpu")
    
    for i, chunk in enumerate(chunks):
        if use_diffusion:
            image = pipe(chunk, num_inference_steps=num_steps).images[0]
            image = image.resize(image_size)
        else:
            image = Image.new("RGB", image_size, (0, 0, 0))
        img_path = f"image_{i}.png"
        image.save(img_path)
        image_paths.append(img_path)
    return image_paths

def create_video(images, durations, speech_path, movie_title, add_subtitles, chunks, image_size=(640, 480)):
    clips = []
    title_clip = mp.TextClip(movie_title, fontsize=50, color='white', size=image_size)
    title_clip = title_clip.set_duration(1).set_position('center')
    black_start = mp.ColorClip(image_size, color=(0,0,0), duration=1).set_opacity(0.8)
    clips.append(mp.CompositeVideoClip([black_start, title_clip]))
    
    for img, dur, chunk in zip(images, durations, chunks):
        frame = np.array(Image.open(img).resize(image_size, Image.Resampling.LANCZOS))
        clip = mp.ImageClip(frame).set_duration(dur)
        if add_subtitles:
            txt_clip = mp.TextClip(chunk, fontsize=30, color='white', size=(image_size[0] - 20, None), method='caption')
            txt_clip = txt_clip.set_duration(dur).set_position(('center', 'bottom'))
            clip = mp.CompositeVideoClip([clip, txt_clip])
        clips.append(clip)
    
    black_end = mp.ColorClip(image_size, color=(0,0,0), duration=2)
    video = mp.concatenate_videoclips(clips + [black_end])
    audio = mp.AudioFileClip(speech_path)
    final_video = video.set_audio(audio)
    final_video.write_videofile("output.mp4", fps=24)
    return "output.mp4"

def process_text(text, movie_title, image_size, use_diffusion, num_steps, add_subtitles):
    chunks = estimate_chunk_durations(text)
    speech_path = generate_speech(text)
    image_paths = generate_images(chunks, image_size, use_diffusion, num_steps)
    durations = [min(10, max(5, len(chunk.split()) / 2.5)) for chunk in chunks]
    video_path = create_video(image_paths, durations, speech_path, movie_title, add_subtitles, chunks, image_size)
    return video_path

with gr.Blocks() as demo:
    gr.Markdown("# Text-to-Video Generator using AI 🎥")
    text_input = gr.Textbox(label="Enter your text")
    movie_title_input = gr.Textbox(label="Movie Title", value="")
    file_input = gr.File(label="Or upload a .txt file")
    image_size_input = gr.Radio(choices=["640x480", "800x600", "1024x768"], label="Select Image Size", value="640x480")
    use_diffusion_input = gr.Checkbox(label="Use Diffusion Images", value=True)
    num_steps_input = gr.Slider(minimum=1, maximum=50, step=1, value=5, label="Diffusion Model Steps")
    add_subtitles_input = gr.Checkbox(label="Add Subtitles", value=False)
    process_btn = gr.Button("Generate Video")
    output_video = gr.Video()
    
    def handle_request(text, movie_title, file, image_size, use_diffusion, num_steps, add_subtitles):
        if file is not None:
            text = open(file.name, "r").read()
        image_size_dict = {"640x480": (640, 480), "800x600": (800, 600), "1024x768": (1024, 768)}
        return process_text(text, movie_title, image_size_dict[image_size], use_diffusion, num_steps, add_subtitles)
    
    process_btn.click(handle_request, inputs=[text_input, movie_title_input, file_input, image_size_input, use_diffusion_input, num_steps_input, add_subtitles_input], outputs=output_video)

demo.launch()