Saltar al contenido principal

Do Not Make the Model Read a Bad Photo

Choose when an image prompt should request a better capture rather than force extraction from a weak photo.

multimodal-ai5 min read

The capture is the problem, not the prompt wording. Priya needs help from an image model. Which prompt is the best first move? Read the exact error code and serial number from this panel. This forces extraction from a weak capture. It violates the readability check: exact text should not be requested until the image is clear enough. Zoom in mentally and infer the likely code from the visible panel layout. This explicitly invites guessing. The violated principle is uncertainty control: do not ask the model to fill unreadable visual gaps. First judge whether the code is readable from this…

Read the full lesson

Sign up free: one personalized lesson every day, matched to your role and goals.

← Back to library

Contáctanos