import gradio as gr
from transformers import AutoTokenizer, VisionEncoderDecoderModel, ViTImageProcessor
from diffusers import StableDiffusionPipeline

# Initialize device and models for captioning
device = 'cpu'
encoder_checkpoint = "nlpconnect/vit-gpt2-image-captioning"
decoder_checkpoint = "nlpconnect/vit-gpt2-image-captioning"
model_checkpoint = "nlpconnect/vit-gpt2-image-captioning"
feature_extractor = ViTImageProcessor.from_pretrained(encoder_checkpoint)
tokenizer = AutoTokenizer.from_pretrained(decoder_checkpoint)
caption_model = VisionEncoderDecoderModel.from_pretrained(model_checkpoint).to(device)

# Load the Stable Diffusion model
diffusion_model = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5")
diffusion_model = diffusion_model.to(device)

def get_caption(image):
    # Generate a caption from the image
    image = image.convert('RGB')
    image_tensor = feature_extractor(images=image, return_tensors="pt").pixel_values.to(device)
    caption_ids = caption_model.generate(image_tensor, max_length=128, num_beams=3)[0]
    caption_text = tokenizer.decode(caption_ids, skip_special_tokens=True)
    return caption_text

def generate_image(caption):
    # Generate an image from the caption
    generated_image = diffusion_model(caption)["sample"][0]
    return generated_image

# Set up Gradio interface
title = "Image Captioning and Generation"
with gr.Blocks(title=title) as demo:
    with gr.Row():
        with gr.Column():
            image_input = gr.Image(label="Upload any Image", type='pil')
            get_caption_btn = gr.Button("Get Caption")
        with gr.Column():
            caption_output = gr.Textbox(label="Caption")
            generate_image_btn = gr.Button("Generate Image")
    with gr.Row():
        generated_image_output = gr.Image(label="Generated Image")

    get_caption_btn.click(get_caption, inputs=image_input, outputs=caption_output)
    generate_image_btn.click(generate_image, inputs=caption_output, outputs=generated_image_output)

demo.launch()