app.py

import cv2
import gradio as gr
import torch
from diffusers import StableDiffusionPipeline
import numpy as np
from transformers.utils import move_cache
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

# Handle Transformers cache migration
move_cache()

# Initialize the Stable Diffusion pipeline
model_id = "CompVis/stable-diffusion-v1-4"
pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)
pipe = pipe.to("cuda")

# Load text summarizer
summarizer_model = "facebook/bart-large-cnn"
tokenizer = AutoTokenizer.from_pretrained(summarizer_model)
summarizer = AutoModelForSeq2SeqLM.from_pretrained(summarizer_model)

# Create video from images using `OpenCV`
def text_to_video(input_text, num_frames=10, fps=2):
    # Summarize the input text
    inputs = tokenizer(input_text, return_tensors="pt", truncation=True)
    summary_ids = summarizer.generate(inputs["input_ids"], max_length=30, min_length=5, length_penalty=2.0)
    prompt = tokenizer.decode(summary_ids[0], skip_special_tokens=True)

    # Generate frames
    frames = []
    for i in range(num_frames):
        prompt_with_frame = f"{prompt}, frame {i+1}"
        image = pipe(prompt_with_frame).images[0]
        frames.append(np.array(image))

    # Save frames as a video
    height, width, layers = frames[0].shape
    video_path = "output.avi"
    out = cv2.VideoWriter(video_path, cv2.VideoWriter_fourcc(*'XVID'), fps, (width, height))

    for frame in frames:
        out.write(cv2.cvtColor(frame, cv2.COLOR_RGB2BGR))
    out.release()
    
    return video_path

# Gradio interface
def generate_video(text, frames, fps):
    video_file = text_to_video(text, num_frames=frames, fps=fps)
    return video_file

interface = gr.Interface(
    fn=generate_video,
    inputs=[
        gr.Textbox(label="Enter your text prompt"),
        gr.Slider(5, 30, value=10, step=1, label="Number of Frames"),
        gr.Slider(1, 10, value=2, step=1, label="Frames per Second (FPS)"),
    ],
    outputs=gr.Video(label="Generated Video"),
    title="Text-to-Video Generator",
    description="Enter a text prompt to generate a short video."
)

if __name__ == "__main__":
    interface.launch()