import gradio as gr
from transformers import pipeline

# Load Hugging Face model pipeline for text generation/paraphrasing
# Using a general-purpose LLM like flan-t5 or bart for paraphrasing
paraphraser = pipeline("text2text-generation", model="Vamsi/T5_Paraphrase_Paws")

# Grammar correction can be handled with a seq2seq grammar model
# or by reprompting paraphraser with grammar-specific instructions
grammar_corrector = pipeline("text2text-generation", model="prithivida/grammar_error_correcter_v1")

def humanize_text(input_text, tone):
    if not input_text.strip():
        return ""

    # Map tone to style instructions
    tone_map = {
        "Natural": "Paraphrase this text in a natural human-like style.",
        "Formal": "Paraphrase this text in a formal professional tone.",
        "Casual": "Paraphrase this text in a casual conversational tone."
    }
    instruction = tone_map.get(tone, tone_map["Natural"])

    # Step 1: Paraphrase with tone
    paraphrased = paraphraser(f"{instruction} Preserve meaning and paragraph breaks. Input: {input_text}",
                               max_length=512, num_return_sequences=1, do_sample=False)[0]['generated_text']

    # Step 2: Grammar correction
    corrected = grammar_corrector(f"Correct grammar and spelling, keep structure: {paraphrased}",
                                  max_length=512, num_return_sequences=1, do_sample=False)[0]['generated_text']

    return corrected

# Gradio UI
demo = gr.Interface(
    fn=humanize_text,
    inputs=[
        gr.Textbox(label="Input Text", lines=10, placeholder="Paste your text here..."),
        gr.Radio(["Natural", "Formal", "Casual"], label="Tone", value="Natural")
    ],
    outputs=gr.Textbox(label="Humanized Output", lines=10),
    title="AI Humanizer",
    description="Humanize AI text into natural, formal, or casual tones while preserving meaning and structure."
)

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