app.py

import gradio as gr
from transformers import BlipProcessor, BlipForConditionalGeneration
from PIL import Image
import torch
from transformers import GPT2Tokenizer, GPT2LMHeadModel

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

processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base").to(device)

tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
gpt2_model = GPT2LMHeadModel.from_pretrained("gpt2").to(device)

def generate_paragraph(image):
    if image.mode != 'RGB':
        image = image.convert('RGB')
    inputs = processor(images=image, return_tensors="pt").to(device)
    output_ids = model.generate(**inputs, max_length=50)
    caption = processor.decode(output_ids[0], skip_special_tokens=True)

    prompt = f"Write a detailed paragraph about this image: {caption}\n\nDetails:"
    tokens = tokenizer.encode(prompt, return_tensors='pt').to(device)
    outputs = gpt2_model.generate(tokens, max_length=150, num_beams=5, no_repeat_ngram_size=2, early_stopping=True, pad_token_id=tokenizer.eos_token_id)
    paragraph = tokenizer.decode(outputs[0], skip_special_tokens=True)

    # Post-process to avoid repeating the prompt
    if paragraph.lower().startswith(prompt.lower()):
        paragraph = paragraph[len(prompt):].strip()

    return paragraph

iface = gr.Interface(
    fn=generate_paragraph,
    inputs=gr.Image(type="pil"),
    outputs="textbox",
    title="Image Paragraph Description Generator",
    description="Upload an image to get a detailed paragraph description generated."
)

iface.launch()