app.py

from transformers import pipeline
from transformers import BlipProcessor, BlipForConditionalGeneration
import streamlit as st
import numpy as np
from PIL import Image
from dotenv import load_dotenv
import openai, os
# load env file 
load_dotenv()

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

api_key=os.getenv("OPENAI_KEY",None)

# image_to_text = pipeline("image-to-text", model="nlpconnect/vit-gpt2-image-captioning")
uploaded_file = st.file_uploader("Choose a Image")
if uploaded_file is not None:
    st.write(uploaded_file)
    image = Image.open(uploaded_file)
    st.image(image, caption='Input', use_column_width=True)
    text = "A photograph of"
    inputs = processor(image, text, return_tensors="pt")
    out = model.generate(**inputs)

    st.write("Caption: ")
    prompt= processor.decode(out[0], skip_special_tokens=True)
    openai.api_key=api_key
    response=openai.Completion.create(
  model="text-davinci-003",
  prompt= 'You are strictly a social media caption generator that absolutely does not include any hashtags and quotation marks. Clearly label the captions "1.", "2." and "3.".\n Generate Creative, Informative and Funny Caption respectively for description : '+prompt,
    temperature=0.6,
    max_tokens=256
)
    

    st.write(response["choices"][0]["text"])