AI Context Exercise: Halloween Display
Instructions
Recreate the Halloween display shown below these instructions using only your own written instructions to AI. You may study the reference image and its numbered list, but do not upload the image, ask AI to describe it, or submit it for copying.
Before you begin, identify the eight items and the specific details that distinguish each one.
Write your initial instructions. Specify appearance, actions, quantities, placement, and exceptions.
Generate the image and compare it carefully with the reference. Check names, counts, colors, shapes, and orientations.
Correct the discrepancies through additional instructions. Be clear about what must change and what must remain.
Submit your final image, your initial instructions, and the corrections you made.
An attractive image alone does not complete the exercise. The goal is to reproduce the specified details accurately.
Please feel free to send me your finished image.
Reasoning
This exercise identifies what has been defined as important and provides the details your instructions must address. In most real-world situations, you must first determine what is important, identify the information needed, and develop the instructions yourself. That preparation is part of the work.
“Create a Halloween scene” leaves most decisions to AI. A blood-red full moon, a witch with Elvira’s face, and three matching tombstones bearing different names require more precise direction.
The fence adds another challenge: most crosses are inverted, but the fourth, eighth, and tenth are upright. Your instructions must establish the general pattern, define its exceptions, and explain where counting begins.
The numbers beneath the exercise image identify separate instruction details. They do not represent a required number of prompts. Several details can be communicated in one prompt, and a clearly stated requirement may still need correction.
This exercise tests observation, preparation, communication, and verification. Do not assume AI will notice missing information, resolve ambiguity as you intended, or preserve every detail during a revision.
Simply copying the image defeats the purpose—and only does you a disservice. The value is in what you learn, not just the image you produce.
What did you learn, and how can you apply it to your next AI task?
Comments