Make a request to a sighted AI for help with a photo (choose one of two cases). Return JSON {"prompt":"...","role":"...","constraints":"..."}: prompt - task for a specific photo, role - what role to give the AI for this case, constraints - restrictions that will make the advice realistic and safe.
An everyday photo is the most “human” scenario: he showed it and asked like an expert friend. But AI only sees what is in the frame and cannot look behind the pipe or touch it. Hence there are two methods. First: give a role - “you are an experienced plumber” or “you are a cook” - the role adjusts the model to the necessary vocabulary and logic, the answer becomes more specific. Second, underrated: ask the AI to ask you clarifying questions if there is not enough data in the photo, instead of guessing. A good assistant will ask “does it drip constantly or when you turn on the water?”, and will not give a diagnosis blindly. For a scenario with food, the hidden lever is the restriction “only from what is visible in the photo, and separately tell me what is missing for the dish”: this way the recipe is realistic, and not “add cream”, which is not there. And always ask for a safety warning (water near the outlet, sharp, hot). A typical rookie mistake: waiting for an accurate diagnosis based on one blurry photo and blindly fixing it according to the first advice.
Unlock access to submit solutions for instant AI review.