app2.py

import os
os.environ.setdefault("GRADIO_ANALYTICS_ENABLED", "False")
import logging
import gradio as gr
from pillow_heif import register_heif_opener

register_heif_opener()

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(img):
  result = captioner(img) 
  caption = result[0]["generated_text"]
  return caption

iface = gr.Interface(
  fn=graptioner,
  inputs=gr.components.Image(type="pil"),
  outputs=["text"],
  allow_flagging="never",
  # analytics_enabled=False
)

iface.launch()