import gradio as gr
import torch
import torch.nn as nn
import torchvision.transforms as transforms
from PIL import Image

# Check for GPU
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# Define the Generator architecture
class Generator(nn.Module):
    def __init__(self, latent_dim=100, img_channels=3, feature_dim=64):
        super(Generator, self).__init__()
        self.latent_dim = latent_dim
        self.model = nn.Sequential(
            nn.ConvTranspose2d(latent_dim, feature_dim * 8, 4, 1, 0, bias=False),
            nn.BatchNorm2d(feature_dim * 8),
            nn.ReLU(True),
            nn.ConvTranspose2d(feature_dim * 8, feature_dim * 4, 4, 2, 1, bias=False),
            nn.BatchNorm2d(feature_dim * 4),
            nn.ReLU(True),
            nn.ConvTranspose2d(feature_dim * 4, feature_dim * 2, 4, 2, 1, bias=False),
            nn.BatchNorm2d(feature_dim * 2),
            nn.ReLU(True),
            nn.ConvTranspose2d(feature_dim * 2, feature_dim, 4, 2, 1, bias=False),
            nn.BatchNorm2d(feature_dim),
            nn.ReLU(True),
            nn.ConvTranspose2d(feature_dim, img_channels, 4, 2, 1, bias=False),
            nn.Tanh()
        )
    
    def forward(self, z):
        return self.model(z)
    
    def generate_latent_space(self, batch_size):
        return torch.randn(batch_size, self.latent_dim, 1, 1, device=device)

# Instantiate the generator and load pre-trained weights
latent_dim = 100
generator = Generator(latent_dim=latent_dim)
# Make sure you have uploaded your pre-trained model file "generator.pth" to your Space
generator.load_state_dict(torch.load("generator.pth", map_location=device))
generator.to(device)
generator.eval()

# Function to generate a face image
def generate_face():
    with torch.no_grad():
        # Generate a random latent vector and produce an image
        z = generator.generate_latent_space(1)
        generated_image = generator(z)
        generated_image = generated_image.cpu().squeeze(0)
        # Denormalize the image (from [-1, 1] to [0, 1])
        generated_image = generated_image * 0.5 + 0.5
        # Convert the tensor to a PIL Image
        to_pil = transforms.ToPILImage()
        image = to_pil(generated_image)
    return image

# Set up the Gradio interface
demo = gr.Interface(
    fn=generate_face,
    inputs=[],  # No inputs – each button press generates a new image
    outputs="image",
    title="CelebA GAN Face Generator",
    description="Generates a face image using a pre-trained GAN on the CelebA dataset.",
)

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