import gradio as gr
import torch
from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler
from diffusers.utils import export_to_video

# Initialize the diffusion pipeline
pipe = DiffusionPipeline.from_pretrained(
    "heboya8/text2video-test-2",    
    torch_dtype=torch.float16,
    # variant="fp16",
    trust_remote_code=True,
)

# Optimize for GPU memory
pipe.enable_model_cpu_offload()
pipe.enable_vae_slicing()

def generate_video(prompt):
    try:
        # Generate video frames
        video_frames = pipe(
            prompt,
            num_inference_steps=1,
            num_frames=1
        ).frames
        
        # Export frames to video file
        video_path = export_to_video(video_frames, output_video_path="output_video.mp4")
        return video_path
    except Exception as e:
        return f"Error generating video: {str(e)}"

# Create Gradio interface
interface = gr.Interface(
    fn=generate_video,
    inputs=gr.Textbox(
        label="Enter your prompt",
        placeholder="e.g., a flower in a garden"
    ),
    outputs=gr.Video(label="Generated Video"),
    title="Text-to-Video Generator",
    description="Enter a text prompt to generate a video using the diffusion model."
)

# Launch the app
interface.launch()