# app.py
import gradio as gr
from transformers import BlipProcessor, BlipForConditionalGeneration
from gtts import gTTS
import io
from PIL import Image

# -------------------------------
# Load BLIP-base model (lighter version)
# -------------------------------
processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base")

# -------------------------------
# Generate caption function
# -------------------------------
def generate_caption_fn(image):
    # Convert uploaded image to PIL
    if not isinstance(image, Image.Image):
        image = Image.fromarray(image)
    
    # BLIP preprocessing
    inputs = processor(images=image, return_tensors="pt")
    
    # Generate caption
    out = model.generate(**inputs)
    caption = processor.decode(out[0], skip_special_tokens=True)
    
    return caption

# -------------------------------
# Convert text to speech using gTTS
# -------------------------------
def text_to_speech(caption):
    tts = gTTS(text=caption, lang='en')
    mp3_fp = io.BytesIO()
    tts.write_to_fp(mp3_fp)
    mp3_fp.seek(0)
    return mp3_fp

# -------------------------------
# Gradio interface: Caption + Audio
# -------------------------------
def generate_caption_tts(image):
    caption = generate_caption_fn(image)
    audio = text_to_speech(caption)
    return caption, audio

interface = gr.Interface(
    fn=generate_caption_tts,
    inputs=gr.Image(type="numpy"),
    outputs=[gr.Textbox(label="Generated Caption"), gr.Audio(type="file", label="TTS Audio")],
    title="Blind Assistant: Image Captioning",
    description="Upload an image and get a descriptive caption + speech."
)

interface.launch()