Add image understanding to my app using MandAPI Chat Completions at https://api.mandapi.com/v1/chat/completions. Read my existing backend and preserve its structure. Ask for an exact account-visible MODEL_ID with current Vision support. Keep MANDAPI_API_KEY on the server and submit text plus image_url content, using a correctly typed data URL for a local image. Start with an offline --demo request preview that makes no network calls. Prepare an explicit --live path for a single user-run image request and show where to inspect usage. State that the recipe is statically adapted and not a previously verified image-inference result. Return a short list of visible objects with uncertainty where appropriate; do not invent detection accuracy or exact measurements.