If you run a cannabis delivery menu, you already know that good product photos and clear descriptions move orders faster than almost anything else. Many small operators are now exploring AI image tools to build menu graphics, banners, and social posts without a full creative budget, and some are tempted to buy midjourney prompts instead of writing every visual brief from scratch. The idea is sound, but the execution needs care, especially in a regulated industry where a single unfortunate image can cause trouble.
Why prompt quality matters more than the tool
Most people who try AI image generators get disappointing results on their first attempts. The tool is rarely the problem. The problem is a vague request like “cool weed picture, modern, green vibes.” A useful prompt specifies composition, lighting, background, color palette, camera angle, and what the image must avoid. When you compare a weak prompt to a structured one, the difference in output is obvious within a few generations.
For a delivery business, that structure matters because you need consistency. Your banners, menu cards, and social posts should look like they belong to the same brand. A reusable prompt template, with fixed slots for lighting style, color family, and layout, lets a non-designer produce a set of images that match each other.
What a reusable template might include
- A fixed brand palette written out in plain color names, not just hex codes, so the tool interprets it consistently
- A lighting and mood descriptor you reuse across every post
- A layout instruction, such as open space on the left for text overlay
- A negative list, specifying elements to exclude like people, packaging close-ups, or anything that looks like a child or a youth-oriented product
Keep notes on which versions worked. Prompt writing is iterative, and a file of tested variations is worth more than a single clever prompt you can’t reproduce next month.
The compliance question comes first
Cannabis advertising is tightly controlled in Washington. The Liquor and Cannabis Board sets rules about what licensed businesses can show and say, including restrictions aimed at keeping advertising away from people under 21 and avoiding content that appeals to minors. Those rules apply to images you generate just as they apply to photos you shoot yourself.
Before you publish any AI-generated image, check it against the questions below. This is a starting checklist, not legal advice, and you should confirm current requirements with the LCB and with a cannabis attorney before launching a campaign.
- Does the image show anyone who could plausibly be under 21?
- Does it use cartoon characters, candy-like shapes, or bright styling that could appeal to children?
- Does it make health, medical, or therapeutic claims?
- Does it depict consumption, or show intoxication in a way that might be prohibited?
- Does the AI output contain text, logos, or trademarks you did not intend to include?
That last point is easy to overlook. Image generators sometimes produce garbled lettering or near-copies of recognizable brand marks. Review every output at full size before it goes anywhere near your menu.
Writing menu copy with prompts
AI text tools can help with product descriptions, but they tend to drift toward medical or exaggerated language. Phrases like “relieves stress” or “cures sleep problems” are exactly the kind of claims you should avoid. Instead, instruct the tool to describe sensory qualities, growing style, or strain category in neutral terms, and then edit the output yourself.
A practical approach is to give the model your verified product data, such as the producer’s name, the product type, the potency as listed on the label, and the test results you have on file, and ask it to write a short description using only those facts. If a detail is not in the data, the prompt should tell the model to leave it out. This keeps the copy tied to what you can actually document.
A simple copy workflow
- Pull the official product information from your inventory system
- Paste it into a prompt that limits the model to those facts and bans health language
- Read the draft against your state-approved wording
- Have a second staff member review before publishing
- Save the approved version so future product listings stay consistent
Buying prompts versus writing your own
Purchased prompt libraries can save time, especially if you are new to image generation. The main advantage is that a good listing shows you what a prompt is designed to produce, along with notes on which tool it was tested in. That context lets you judge whether it fits your brand before you spend anything.
The risk is assuming a purchased prompt is compliant simply because someone else wrote it. Prompts written for fashion, travel, or food photography will often produce imagery that works poorly or violates cannabis rules when adapted. Treat any outside prompt as a draft. Adjust the wording to remove people, add your required negative terms, and run the same compliance review you would apply to your own work.
If you want to see how structured prompt listings are organized before you commit to any approach, look through the examples in a dedicated prompt marketplace for tested AI creative prompts. Pay attention to how each entry explains its intended use, since that habit translates well into writing your own internal prompt library.
Building an internal prompt library
Whether you buy prompts or write them, the long-term asset is your own library. A simple spreadsheet works well. Record the prompt text, the tool and version it was run in, the intended use such as Instagram story, menu header, or delivery email, and the date it was reviewed for compliance. Add a column for the outcome so you can see which versions consistently produce usable results.
Review the library every few months. Advertising rules can change, product lines rotate, and AI tools update their models in ways that shift output. A prompt that produced clean results last year may need rewording now.
Keep humans in the loop
AI tools are good at producing options quickly, but they are poor at judging context. They do not know your neighborhood, your customer base, or what a particular local event means to your community. A staff member who knows your delivery area should make the final call on every image and every line of copy. Treat the tool as a fast first draft, not a publisher.
For a small delivery business in Pierce County, the realistic goal is not flashy AI campaigns. It is a steady, compliant flow of clear menu graphics and honest product descriptions that help customers find what they need. Used carefully, prompts can make that workload manageable without putting your license at risk.
Leave a Reply