When our team at Kailua Marijuana Delivery started experimenting with AI image tools, we quickly learned that the tool mattered less than the prompt. Two people using the same generator could produce wildly different results, and the difference usually came down to how clearly the request was written. If you are tempted to skip the trial-and-error phase, you can buy midjourney prompts that other creators have already tested, then adapt them to your own brand. That shortcut only pays off if you judge the output honestly and keep your compliance obligations in front of you the whole time.
Why prompts matter more than the generator
Most image generators will respond to almost anything, which is exactly the problem. A vague request like ‘cannabis shop, nice vibe’ gives you generic stock-photo energy. A specific request that names the setting, light source, lens, color palette, and mood gives you something you can actually use on a menu board or a social post. The prompt is effectively your creative brief, and a weak brief wastes time no matter how powerful the model is.
What we actually needed images for
Before writing anything, we listed the jobs we wanted the imagery to do. That list kept us from generating pretty pictures with no purpose. For our delivery business, the main needs were:
- Header art for the blog and email newsletter
- Seasonal scenes tied to the neighborhood, such as a quiet morning near Kalama Beach or an evening on a lanai with trade winds in the palms
- Concept mockups for label and packaging layouts that would be reviewed by a human before any real printing
- Background textures for a printed delivery insert
Each job had different constraints. A newsletter header can be atmospheric. A packaging concept needs clean space for required text, which is why we started with prompts that asked for negative space on one side of the frame.
Compliance comes first, not last
Cannabis advertising is regulated, and the rules that apply to your business depend on your state, your county, and the platforms you use. We are not lawyers, and nothing in this article replaces advice from a qualified attorney or your licensing authority. That said, these are the principles we apply to every AI-generated image before it leaves our desk:
- No people who appear to be minors, and no playful cartoon styles that could appeal to children
- No depictions of consumption that look like an endorsement or that show intoxication as desirable
- No imagery that implies medical outcomes, cures, or health benefits
- No real trademarks, logos, or celebrity likenesses that the model may reproduce by accident
- A human review step for every image, with a written checklist
We also add a guardrail line to our prompt templates that asks the model to avoid text, logos, and recognizable faces. Models do not always obey, so the review step is still mandatory.
How we judge a prompt before using it
When evaluating any prompt, whether we wrote it or bought it, we run it at least three times and compare the outputs. We look for consistency more than one lucky image. A good prompt should produce recognizably similar results across runs, with variation that you can control through small edits.
- Does the output match the brief without extra objects we did not ask for?
- Is there clear space for text if the image is headed to print or a banner?
- Does the lighting feel natural for Hawaii, with warm late-afternoon tones rather than neon?
- Would a customer see anything that could be read as a health claim or youth appeal?
- Can we reproduce the look with a small tweak, such as changing only the time of day?
Comparing options in a structured library
Once we had a handful of prompts that worked, the next problem was organization. Notes scattered across phone apps and shared documents made it hard to see what we already had. We eventually found it useful to review a curated prompt marketplace where each listing shows example outputs, which let us compare styles side by side before deciding what to adapt. Seeing the example image next to the prompt text made it much easier to understand which words were doing the real work.
Building your own prompt library
Whether you buy prompts or write them yourself, you will eventually want your own library. We use a simple template with fixed fields so that anyone on the team can pick up a prompt and get consistent results:
- Subject: what is in the frame
- Setting: location and time of day, with local details
- Lighting: direction, color temperature, and softness
- Lens and framing: wide, close, or overhead, plus where the negative space sits
- Palette: three to five colors that match our brand guide
- Mood: calm, energetic, or nostalgic, kept in a single word
- Guardrails: no text, no logos, no minors, no consumption imagery
Keeping the guardrail line in every template is the single most useful habit we have built. It turns compliance from a memory test into a default.
Common mistakes we made
- Overloading one prompt with too many ideas, which produced cluttered images
- Trusting the first output and skipping the review checklist
- Changing the prompt and the model at the same time, which made results impossible to compare
- Forgetting that image rights and licensing terms can differ between tools and plans
- Using seasonal imagery that did not match the actual season in Kailua, which customers noticed quickly
Final thoughts for local delivery businesses
AI image tools can save real time for a small cannabis delivery team, but they reward discipline. Write clear briefs, keep compliance rules visible, review every output with a human eye, and build a library that your whole team can use. A neighborhood business succeeds when its visuals feel like they belong to the place it serves, and that comes from local detail and careful editing rather than from any single magic prompt. Start small, test honestly, and let your customers in Kailua tell you what feels authentic.

Leave a Reply