Create app.py

app.py

@@ -0,0 +1,57 @@
+import gradio as gr
+import torch
+from diffusers import StableDiffusionPipeline
+from moviepy.editor import ImageSequenceClip
+import os
+import numpy as np
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+
+# Initialize text-to-image pipeline
+model_id = "CompVis/stable-diffusion-v1-4"
+pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)
+pipe = pipe.to("cuda")
+
+# Load a text summarization model for better prompts (optional)
+summarizer_model = "facebook/bart-large-cnn"
+tokenizer = AutoTokenizer.from_pretrained(summarizer_model)
+summarizer = AutoModelForSeq2SeqLM.from_pretrained(summarizer_model)
+
+# Function to create video from text
+def text_to_video(input_text, num_frames=10, fps=2):
+    # Summarize input text for better image prompts
+    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 images
+    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 images as video
+    video_path = "output.mp4"
+    clip = ImageSequenceClip(frames, fps=fps)
+    clip.write_videofile(video_path, codec="libx264")
+    return video_path
+
+# Define 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()