app.py

import torch
from PIL import Image
import gradio as gr
from transformers import Blip2Processor, Blip2ForConditionalGeneration, pipeline
from gtts import gTTS
import os

device = "cuda" if torch.cuda.is_available() else "cpu"

# Load BLIP-2
processor = Blip2Processor.from_pretrained("Salesforce/blip2-opt-2.7b")
model = Blip2ForConditionalGeneration.from_pretrained(
    "Salesforce/blip2-opt-2.7b",
    torch_dtype=torch.float16 if device == "cuda" else torch.float32
).to(device)

# Load Whisper pipeline for speech-to-text
whisper_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base")

# Store image globally
current_image = {"image": None}

def load_image(image):
    current_image["image"] = image
    return "Image uploaded. Now ask a question via voice."

def ask_question(audio):
    if current_image["image"] is None:
        return "Please upload an image first.", None

    # Transcribe speech
    question = whisper_pipe(audio)["text"]

    # Ask BLIP-2
    inputs = processor(current_image["image"], question, return_tensors="pt").to(device, torch.float16 if device == "cuda" else torch.float32)
    output = model.generate(**inputs, max_new_tokens=100)
    answer = processor.decode(output[0], skip_special_tokens=True)

    # Convert to speech
    tts = gTTS(answer)
    tts.save("answer.mp3")

    return f"Q: {question}\nA: {answer}", "answer.mp3"

# Gradio UI
with gr.Blocks() as app:
    gr.Markdown("# 🧠🖼️ Ask-the-Image with BLIP-2 + Whisper + gTTS")
    with gr.Row():
        image_input = gr.Image(type="pil", label="Upload Image")
        image_status = gr.Textbox(label="Status", interactive=False)

    audio_input = gr.Audio(source="microphone", type="filepath", label="Ask a Question (voice)")
    output_text = gr.Textbox(label="Q&A", lines=4)
    output_audio = gr.Audio(label="Answer (speech)")

    image_input.change(fn=load_image, inputs=image_input, outputs=image_status)
    audio_input.change(fn=ask_question, inputs=audio_input, outputs=[output_text, output_audio])

app.launch()