app.py

import os
import cohere
import gradio as gr
from transformers import AutoTokenizer, ViTFeatureExtractor, VisionEncoderDecoderModel

COHERE_API_KEY = os.getenv('COHERE_API_KEY')

co_client = cohere.Client(COHERE_API_KEY)

device = 'cpu'
encoder_checkpoint = "nlpconnect/vit-gpt2-image-captioning"
decoder_checkpoint = "nlpconnect/vit-gpt2-image-captioning"
model_checkpoint = "nlpconnect/vit-gpt2-image-captioning"
feature_extractor = ViTFeatureExtractor.from_pretrained(encoder_checkpoint)

tokenizer = AutoTokenizer.from_pretrained(decoder_checkpoint)
model = VisionEncoderDecoderModel.from_pretrained(model_checkpoint).to(device)


def predict(image):
    """Predict the generic image caption from the image """
    # image = image.convert('RGB')
    image = feature_extractor(image, return_tensors="pt").pixel_values.to(device)
    clean_text = lambda x: x.replace('<|endoftext|>', '').split('\n')[0]
    caption_ids = model.generate(image, max_length=125)[0]
    img_caption_text = clean_text(tokenizer.decode(caption_ids))
    
    caption_text = creative_caption(img_caption_text)
    hashtags = caption_hashtags(img_caption_text)
    
    return caption_text, hashtags


def creative_caption(text):
    return co_client.generate(prompt=f"Write some trendy instagram captions for the following prompt - {text}").generations[0].text


def caption_hashtags(text):
    return co_client.generate(prompt=f"Write some trendy instagram hashtags for the following prompt - {text}").generations[0].text


input_upload = gr.Image(label="Upload any Image")
output = [
    gr.Textbox(label="Captions"),
    gr.Textbox(label="Hashtags"),
]

title = "Instagram Image Captioning"
description = "Made for Linesh"
interface = gr.Interface(

    fn=predict,
    description=description,
    inputs=input_upload,
    theme="grass",
    outputs=output,
    title=title,
)
interface.launch(debug=True)