app.py

# 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-large")
model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-large")

# -------------------------------
# Generate caption function
# -------------------------------
# def generate_caption_tts(image):
#     caption = generate_caption(model, processor, image)
#     audio_file = text_to_audio_file(caption)
#     return caption, audio_file  # return file path, not BytesIO


# -------------------------------
# Convert text to speech using gTTS
# -------------------------------
import tempfile
import pyttsx3

def text_to_audio_file(text):
    # Create a temporary file
    tmp_file = tempfile.NamedTemporaryFile(suffix=".mp3", delete=False)
    tmp_path = tmp_file.name
    tmp_file.close()

    engine = pyttsx3.init()
    engine.save_to_file(text, tmp_path)
    engine.runAndWait()

    return tmp_path

def generate_caption_from_image(model, processor, image):
    # image: PIL.Image
    inputs = processor(images=image, return_tensors="pt")
    out = model.generate(**inputs)
    caption = processor.decode(out[0], skip_special_tokens=True)
    return caption
# -------------------------------
# Gradio interface: Caption + Audio
# -------------------------------
def generate_caption_tts(image):
    caption = generate_caption_from_image(model, processor, image)  # uses global model/processor
    # audio_file = text_to_audio_file(caption)
    return caption 



interface = gr.Interface(
    fn=generate_caption_tts,
    inputs=gr.Image(type="numpy"),
    outputs=[gr.Textbox(label="Generated Caption")],
    title="Image Captioning for Visually Impaired",
    description="Upload an image, get a caption and audio description."
)


interface.launch()
# # demo.launch(share=True)



# import gradio as gr
# from transformers import AutoProcessor, AutoModelForCausalLM
# import torch
# from PIL import Image

# # Load small LLaVA model
# processor = AutoProcessor.from_pretrained("llava/LLaVA-7B-llm-small")
# model = AutoModelForCausalLM.from_pretrained(
#     "llava/LLaVA-7B-llm-small",
#     torch_dtype=torch.float16,
#     device_map="auto"  # Automatically use GPU if available
# )

# def generate_caption(image):
#     # Convert to PIL if needed
#     if isinstance(image, str):
#         image = Image.open(image).convert("RGB")
    
#     # Prepare inputs
#     inputs = processor(images=image, return_tensors="pt").to(model.device)
    
#     # Generate output
#     outputs = model.generate(**inputs, max_new_tokens=50)
    
#     # Decode result
#     caption = processor.decode(outputs[0], skip_special_tokens=True)
#     return caption

# # Gradio Interface
# interface = gr.Interface(
#     fn=generate_caption,
#     inputs=gr.Image(type="pil"),
#     outputs=gr.Textbox(label="Generated Caption"),
#     title="LLaVA Image Captioning"
# )

# interface.launch()