If your team has ever asked an AI image tool for a menu banner and received something that looked nothing like your brand, you already know the difference between a casual request and a usable one. Many operators start by searching for midjourney prompts and quickly realize that a prompt copied from a forum rarely fits a delivery business with its own products, service area, and legal constraints. The real opportunity is building a small, tested library of prompts that your team understands and can reuse.
Why generic prompts fail delivery shops
A delivery menu is not a lifestyle magazine. Customers scan it on a phone between meetings, often looking for a specific strain type, a edible dose format, or a reliable delivery window. A prompt that asks for a ‘dreamy, cosmic cannabis scene’ may produce striking art, but it tells the model nothing about layout, aspect ratio, or the text that must remain readable at thumbnail size.
Generic prompts also tend to drift toward imagery that creates compliance problems. Cartoon styles, bright candy packaging, and anything that resembles children’s products can attract scrutiny, and a model left unguided will sometimes produce them without warning.
What makes a prompt actually work
Reliable prompts share a handful of traits. They define the output format before they describe the aesthetic. They specify constraints the model must respect, and they say what to avoid. Working through dozens of iterations at a small shop, the pattern that holds up is this:
- Purpose: the exact place the asset will appear, such as an Instagram story, a homepage hero, or an email header.
- Dimensions and safe zones: the aspect ratio plus a note about where text overlays will sit.
- Style anchors: two or three concrete references, like matte packaging photography or flat editorial illustration, rather than adjectives alone.
- Negative constraints: what must never appear, such as people consuming product, cartoon characters, or anything that looks like a toy or candy.
- Brand variables: placeholders for color hex codes, product names, and offer language so the same template works across campaigns.
Compliance comes first
Cannabis marketing is regulated, and rules differ by state and locality. Colorado has detailed requirements about advertising content, including restrictions on material that appeals to people under 21 and rules on accurate, non-misleading claims. Before any prompt goes into production, have your compliance advisor or attorney review the categories of imagery and copy you plan to generate. A prompt library is only as good as the guardrails written into it.
Practical guardrails for your prompts include:
- Ban cartoon, animated, or mascot-style characters in every template.
- Ban depictions of consumption, intoxication, or minors in any setting.
- Never include health, medical, or therapeutic claims in generated copy. Words like ‘cures’ or ‘treats’ should be blocked by default.
- Keep product photography realistic and tied to your actual packaging, not invented packaging.
- Require age-gate language and licensing information on any template used for web or print.
Treat these guardrails as part of the prompt itself, not as an afterthought. A line like ‘no characters, no consumption, no health claims, realistic packaging only’ belongs in every template header.
Five prompt categories worth building first
1. Menu hero banners
Build one template for your homepage hero and one for category pages. Specify the crop, the empty space reserved for headline text, and a restrained color palette that matches your logo. Generate three variants per campaign and pick the one that keeps the most negative space.
2. Product card descriptions
Text prompts matter as much as image prompts. A good product copy template takes the verified lab results, the product category, and the customer-facing benefits you are permitted to state, then outputs a short description under a fixed character count. Forcing the model to work from supplied facts reduces invented details.
3. Email subject lines and preview text
Ask for ten options in a fixed format, each under 45 characters, with no urgency tricks like false countdowns. Your team then picks and tests. This is a low-risk place to use AI because a human reviews every line before sending.
4. Delivery-day social graphics
Create a template for weekend delivery windows or new arrivals that uses your brand grid. Keep it simple: one product photo, one headline, one call to action pointing to the menu.
5. Seasonal and event artwork
Holidays and local Denver events are good prompt triggers, but keep the imagery abstract or landscape-based. Mountain light, city skylines, and neutral textures work better than anything that could be read as a party scene involving consumption.
A workflow your team can actually run
Prompts degrade without maintenance, so assign ownership. One person writes or adapts templates, a second reviews for compliance, and a third logs results. A simple shared document with columns for prompt version, date, channel, and outcome is enough to start. If you want a starting point rather than a blank page, a prompt marketplace built around tested, reusable templates can give your writers a structure to adapt to your brand and rules instead of building everything from scratch.
Version everything
When a prompt produces a strong result, save the exact text, the model version, and the settings. When it fails, note what went wrong in one sentence. Over a few months this log becomes more valuable than any single template because it reflects your actual customers and your actual legal environment.
Review before publishing, every time
No AI output goes live without a human check for accuracy, compliance, and brand fit. Images should be checked for stray text, distorted packaging, and anything that accidentally reads as a child-oriented product. Copy should be checked against your approved claims list.
Measuring results without guessing
It is tempting to judge prompts by how impressive the output looks. A better measure is whether the asset does its job. Track click-through rates on menu links, unsubscribes on email campaigns, and how often a template needs significant manual editing. A prompt that needs heavy correction is not saving time, even if the first draft looks polished. Use your own analytics to set baselines before you change anything, then compare after each revision.
The takeaway
AI prompts can help a cannabis delivery business produce consistent menu art, clearer product copy, and faster campaign turnaround, but only when they are specific, tested, and bounded by compliance rules. Start with two or three templates, write the guardrails into every one, keep a simple log of what works, and make human review non-negotiable. Over time you will have a prompt library that reflects your brand and your legal obligations, which is far more useful than any collection of generic examples.
Leave a Reply