app.py

import logging
import os

import gradio as gr
from pillow_heif import register_heif_opener

register_heif_opener()

import gradio as gr
from transformers import pipeline

LOG_LEVEL = os.getenv("LOG_LEVEL", "DEBUG")
MAX_NEW_TOKENS = int(os.getenv("MAX_NEW_TOKENS", 200))
# https://huggingface.co/models?pipeline_tag=image-to-text&sort=likes
MODEL = os.getenv("MODEL", "Salesforce/blip-image-captioning-large")

logging.basicConfig(level=LOG_LEVEL)
logger = logging.getLogger(__name__)


logger.info("Loading model...")
# simpler model: "ydshieh/vit-gpt2-coco-en"
captioner = pipeline(
    "image-to-text",
    model=MODEL,
    max_new_tokens=MAX_NEW_TOKENS,
)
logger.info("Done loading model.")


def graptioner(image_url):
    global captioner
    result = captioner(image_url)
    caption = result[0]["generated_text"]
    return caption


# add gradio interface
iface = gr.Interface(fn=graptioner, inputs="text", outputs=["text"], allow_flagging="never")
iface.launch()