ask.py

"""Visual Q&A — up to 5 questions about one file, priced like a single extraction.

    python examples/ask.py photo.jpg "Is there a dog in the image?" "How many people?"
"""

import sys

from visionapi import VisionAPI

if len(sys.argv) < 2:
    raise SystemExit("usage: python examples/ask.py <file> [question…]")

path, *questions = sys.argv[1:]
vision = VisionAPI()

res = vision.ask(file=path, questions=questions or ["What is in this image?"])

for a in res["answers"]:
    # `verdict` is the machine-readable half: yes | no | uncertain | n/a.
    # "uncertain" is a yes/no question the image does not settle — a real answer, not a
    # failure. "n/a" means it wasn't a yes/no question in the first place.
    print(f"Q: {a['question']}")
    print(f"A: {a['answer']}")
    print(f"   verdict={a['verdict']} confidence={a['confidence']}\n")

print(
    f"{res['credits_used']} credit(s) — the questions themselves are free, "
    "you pay per image or per selected page."
)