Our team at Hilo Cannabis Delivery spent several months experimenting with image generators for seasonal menus, social posts, and holiday banners. The early results were disappointing. Many of the pictures looked like generic stock photography, with glossy palm trees and nothing that felt like Hilo. When we started looking at how other creators handled this, we found a growing number of people sharing tested wording through ai art prompts that are written to be reused and refined. That was the moment we realized the problem was not the tools. It was the briefs we were giving them.
Why most prompts produce forgettable output
A typical first attempt looks something like this: “beautiful cannabis plant, modern style, high quality.” The model has to guess almost everything. It guesses the lighting, the setting, the mood, and the level of detail. What comes back is an average of everything it has seen, which is why so many results look interchangeable.
Specific prompts fix this. Compare the earlier example with something like: “overhead photo of a woven lauhala mat on a weathered koa table, morning light through a window, a ceramic mug and a notebook with handwritten grocery list, soft shadows, muted greens and warm wood tones, shot on a 50mm lens.” The second version gives the model a place, objects, light direction, color palette, and camera choice. Each of those is a decision the model no longer has to make on its own.
A simple structure for briefs that hold up
After a lot of trial and error, we settled on a five-part structure that we now use for every creative request:
- Subject: the main thing in the frame, named concretely.
- Setting: a real location or texture, such as black volcanic rock, a rain-soaked lanai, or the slope of Mauna Kea at dusk.
- Light: time of day, source, and quality, such as late afternoon sun through trade-wind clouds.
- Palette: three to five colors you actually want, not “vibrant.”
- Constraints: what to avoid, such as text in the image, visible faces, or branded packaging.
The constraints line matters more than people expect. For a regulated product, we always tell the model to leave out anything that could look like it targets young people, such as cartoon characters, candy-like packaging, or school settings. Keeping that rule inside the prompt is far easier than fixing it in post.
Making the place feel like Hilo
Generic prompts are a big part of why local businesses end up with lookalike content. Hilo has specific visual cues that are easy to describe and hard for a model to guess: the rain that comes in sideways off Hamakua, the banyan roots on Waianuenue Avenue, the way the bay turns from gray to copper in the evening, and the dark sand and clinkers near the coast. When we include two or three of these in a brief, the output stops looking like it could be from any tropical island.
We also write captions in the same specific way. Instead of “enjoy our products this weekend,” we describe what a customer would actually see when a driver pulls up: a box with a receipt, a note about parking near the Kaumana side, or a reminder that the order was verified at the door.
Compliance is part of the creative process
Cannabis marketing carries rules that a general creative team might not know about. Before publishing any AI-generated image, we check it against current state and county requirements, which change over time, so we confirm them directly with the relevant agencies rather than relying on memory. In practice, our review checklist covers: To go deeper, explore The marketplace for AI prompts that actually work.
- No health, healing, or medical claims in images or captions.
- No depictions of consumption or people who appear under the legal age.
- No imagery that mimics snacks, candy, or beverages.
- Age-gate language present wherever the rules require it.
- A human reviewer signs off on every final file.
AI tools are good at making things look appealing, which is exactly why a human check matters. A picture can be attractive and still be noncompliant, and the responsibility stays with us as the business.
Building a small prompt library
We now keep a shared document with prompts that worked, along with notes on what changed the result. Each entry records the model version, the prompt, the parameters, and one line on why it succeeded or failed. Over time this has saved us more hours than any single tool upgrade. A new team member can start from a tested brief instead of a blank page.
If you are building something similar, start small. Pick one recurring need, such as a weekly menu header or a delivery-zone map graphic. Write five versions of the brief, change one element at a time, and keep the one that works. Treat the process like product testing, because that is essentially what it is.
What we would do differently
Looking back, we wasted the first two months chasing a single perfect prompt. We should have focused on consistency instead. A set of reusable brief templates, a clear compliance checklist, and a habit of writing down what worked would have gotten us further, faster. The creative output is only as good as the thinking that goes into the request.
For a delivery business on the Big Island, the goal is simple: content that feels like home to local customers and stays clearly within the rules. Specific prompts, a steady review process, and real Hilo details get us most of the way there.

Leave a Reply