import gradio as gr
import replicate
import os
import random
import openai
import numpy as np
from PIL import Image
import requests
import io
import base64
import zipfile
from transformers import pipeline

# Set API tokens
os.environ["REPLICATE_API_TOKEN"] = "r8_Brv0MtpmAiqrXrMrziyUXoSHuFV5hqs1Lw4Mo"
# Initialize the Replicate client
rep_client = replicate.Client()

# Set your OpenAI API key
OPENAI_API_KEY = "sk-proj-5iy4bwrqAW8GpguiEawaT3BlbkFJ8p88lLSjOCeDbxWsAOlr"
openai.api_key = OPENAI_API_KEY

# Load sentiment analysis model
sentiment_analysis = pipeline('sentiment-analysis')

predefined_prompts = [
    "Missing bolts on railway track",
    "Cracks on railway track",
    "Overgrown vegetation near railway track",
    "Broken railings on railway bridge",
    "Debris on railway track",
    "Damaged railway platform"
]

def analyze_feedback(feedback):
    result = sentiment_analysis(feedback)[0]
    sentiment = result['label']
    score = result['score']
    if sentiment == "POSITIVE":
        return f"Thank you for your positive feedback! Your satisfaction score is {score}."
    else:
        return f"Sorry to hear that. We are trying to improve based on your feedback. Your dissatisfaction score is {score}."

def generate_variations(base_prompt, number_of_variations):
    locations = ["on the left side", "on the right side", "at the top", "at the bottom", "in the center"]
    sizes = ["small", "medium", "large", "tiny", "huge"]
    weather_conditions = ["under cold conditions", "during hot weather", "in dry weather", "in humid conditions", "under varying temperatures"]

    variations = []
    for _ in range(number_of_variations):
        location = random.choice(locations)
        size = random.choice(sizes)
        weather = random.choice(weather_conditions)
        
        full_prompt = f"{base_prompt}, with a {size} defect {location}, observed {weather}."
        variations.append(full_prompt)
    return variations

def generate_images(prompts):
    images = []
    for prompt in prompts:
        try:
            prediction = rep_client.predictions.create(
                version="ac732df83cea7fff18b8472768c88ad041fa750ff7682a21affe81863cbe77e4",
                input={"prompt": prompt, "scheduler": "K_EULER"}
            )
            prediction.wait()
            if prediction.status == "succeeded" and prediction.output:
                images.append(prediction.output[0])
            else:
                images.append("Failed to generate image.")
        except Exception as e:
            images.append(f"Error: {str(e)}")
    return images

# UI creation
with gr.Blocks() as app:
    with gr.Tabs("Prompt Input"):
        with gr.Tab("Generate Images"):
            prompt_input = gr.Dropdown(choices=predefined_prompts, label="Select a defect prompt")
            number_input = gr.Number(label="Number of images", value=1, minimum=1, maximum=10)
            generate_button = gr.Button("Generate")
            gallery = gr.Gallery(label="Generated Images")
            
            generate_button.click(
                fn=lambda prompt, num: generate_images(generate_variations(prompt, num)),
                inputs=[prompt_input, number_input],
                outputs=gallery
            )

        with gr.Tab("Custom Defect"):
            custom_prompt_input = gr.Textbox(label="Custom Defect")
            number_input_custom = gr.Number(label="Number of images to generate", value=1, minimum=1, maximum=10)
            submit_button_custom = gr.Button("Generate")
            image_outputs_custom = gr.Gallery()

            submit_button_custom.click(
                fn=lambda prompt, num: generate_images(generate_variations(prompt, num)),
                inputs=[custom_prompt_input, number_input_custom],
                outputs=image_outputs_custom
            )
            
    feedback_input = gr.Textbox(label="Enter your feedback", placeholder="Write your feedback here...")
    like_button = gr.Button(value="👍 Like")
    dislike_button = gr.Button(value="👎 Dislike")
    feedback_result = gr.Textbox(label="System Response", interactive=False)
    refresh_button = gr.Button("Refresh Page")
    

    like_button.click(lambda x: analyze_feedback(x), inputs=feedback_input, outputs=feedback_result)
    dislike_button.click(lambda x: analyze_feedback(x), inputs=feedback_input, outputs=feedback_result)
    refresh_button.click(lambda: gr.update(reload_browser=True))

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