from fastai.vision.all import *
import gradio as gr
import fal_client
from PIL import Image
import io
import random
import requests
from pathlib import Path

# Load your model
learn = load_learner('resnet50_30_categories.pkl')

# Wikipedia links
search_terms_wikipedia = {
    "blazing star": "https://en.wikipedia.org/wiki/Mentzelia",
    "bristlecone pine": "https://en.wikipedia.org/wiki/Pinus_longaeva",
    # ... (same as before)
    "goldfields coreopsis": "https://en.wikipedia.org/wiki/Coreopsis"
}

# Prompt templates for art generation
prompt_templates = [
    "A dreamy watercolor scene of a {flower} on a misty morning trail, with golden sunbeams filtering through towering redwoods, and a curious hummingbird hovering nearby.",
    "A loose, expressive watercolor sketch of a {flower} in a wild meadow, surrounded by dancing butterflies and morning dew drops sparkling like diamonds in the dawn light.",
    "An artist's nature journal page featuring a detailed {flower} study, with delicate ink lines and soft watercolor washes, complete with small sketches of bees and field notes in the margins.",
    "A vibrant plein air painting of a {flower} patch along a coastal hiking trail, with crashing waves and rugged cliffs in the background, painted in bold, energetic brushstrokes.",
    "A whimsical mixed-media scene of a {flower} garden at sunrise, combining loose watercolor washes with detailed botanical illustrations, featuring hidden wildlife and morning fog rolling through the valley."
]

# Example image paths (replace with actual paths on your system or Hugging Face space)
example_images = [
    str(Path('example_images/example_1.jpg')),
    str(Path('example_images/example_2.jpg')),
    str(Path('example_images/example_3.jpg')),
    str(Path('example_images/example_4.jpg')),
    str(Path('example_images/example_5.jpg'))
]

def on_queue_update(update):
    if isinstance(update, fal_client.InProgress):
        for log in update.logs:
            print(log["message"])
    else:
        print("Received non-InProgress update:", update)

# Processing function
def process_image(img):
    predicted_class, _, probs = learn.predict(img)
    classification_results = dict(zip(learn.dls.vocab, map(float, probs)))

    # Wikipedia
    wiki_url = search_terms_wikipedia.get(predicted_class.lower(), "No Wikipedia entry found.")

    # Generate image via FAL
    result = fal_client.subscribe(
        "fal-ai/flux/schnell",
        arguments={
            "prompt": random.choice(prompt_templates).format(flower=predicted_class),
            "image_size": "portrait_4_3"
        },
        with_logs=True,
        on_queue_update=on_queue_update,
    )

    image_url = result['images'][0]['url']
    response = requests.get(image_url)
    generated_image = Image.open(io.BytesIO(response.content))

    return classification_results, generated_image, wiki_url

# Interface
with gr.Blocks() as demo:
    gr.Markdown("# 🌼 Wildflower Classifier & Artistic Generator")

    with gr.Row():
        input_image = gr.Image(height=230, width=230, label="Upload an image", type="pil")

    with gr.Row():
        with gr.Column():
            label_output = gr.Label(label="Prediction")
            wiki_output = gr.Textbox(label="Wikipedia Link")
        generated_image = gr.Image(label="AI Artistic Interpretation")

    gr.Examples(
        examples=example_images,
        inputs=input_image,
        examples_per_page=6
    )

    input_image.upload(
        fn=process_image,
        inputs=input_image,
        outputs=[label_output, generated_image, wiki_output]
    )

demo.launch()