If your team has ever typed a request into an image generator and received something that looks nothing like the weekend delivery banner you pictured, the problem is rarely the tool. It is usually the wording. Many small operators are discovering that the gap between a useless render and a usable graphic comes down to how the request is structured, and a growing number of creators now study midjourney prompts to see how working instructions are put together before writing their own.
Why generic AI art misses the mark for delivery brands
Cannabis delivery sits in an unusual visual space. Your customers want to see clean packaging, friendly courier services, and neighborhood context, but the imagery you can legally and ethically publish is narrow. A vague request like “cannabis delivery van, modern, green” tends to produce cartoonish leaves, cliched smoke clouds, or vehicles with no recognizable branding. Those results are not just ugly. They can also send the wrong message about the kind of business you run.
Specific prompts fix most of this. When you describe the scene the way a photographer would brief a shoot, the output becomes far more predictable.
What a usable prompt contains
Across the prompts that tend to hold up over multiple revisions, a few components show up consistently:
- Subject: the main object or person, described concretely (a white unmarked hatchback parked at a curb, a courier in a dark jacket holding a sealed paper bag).
- Setting: the place and time of day, such as a residential street in late afternoon or a quiet apartment lobby.
- Lighting and lens: soft window light, 35mm framing, shallow depth of field. These words do more to set the mood than adjectives like “premium” or “luxury.”
- Brand constraints: your exact color hex codes, typography direction, and logo placement zone, even if you add the logo manually afterward.
- Exclusions: a plain list of things you do not want, such as people consuming products, cartoon characters, or readable third-party brand names.
Writing exclusions explicitly matters more than most people expect. Without them, generators often add elements that you would never approve on a second look.
Compliance-first prompting
Cannabis advertising is regulated at the state and municipal level, and platform policies on social media, search, and map listings can be stricter still. Treat your prompt library as part of your compliance process, not as a creative shortcut. Before any generated image goes live, check it against your local rules and your ad platform guidelines. This article is general information, not legal advice, so confirm details with a qualified advisor in your jurisdiction.
Practical guardrails many delivery operators build into their prompts include:
- No depiction of anyone consuming cannabis or appearing intoxicated.
- No characters, animals, or art styles that could appeal to children.
- No health, medical, or therapeutic claims in any text the generator adds to an image.
- No fake product packaging. If you show a package, use your actual approved label, added in design software after generation.
- No implied location claims, such as suggesting you deliver to areas where you are not licensed to operate.
Keep these rules in a shared document so that a new team member writing a prompt at 11 p.m. follows the same standards as the owner.
Finding and testing prompts before you rely on them
You do not need to write every prompt from scratch. Searching for examples is a reasonable starting point, but the quality varies widely. A prompt that produced a stunning result for someone else may fail for your color palette or aspect ratio. Look for marketplaces where listings show sample outputs, describe the intended use, and include notes on known limitations. A prompt marketplace that lets buyers review listings before purchase can save time, but the same discipline applies: run every purchased prompt through your own brand and compliance checks before it touches a public channel.
When testing, change one variable at a time. Swap the setting, then the lighting, then the framing. You will learn quickly which words the generator responds to and which it ignores.
Building a prompt library for your menu and promotions
A small, well-organized library beats a folder of hundreds of untested experiments. Consider grouping prompts by use case:
- Hero banners: wide, uncluttered scenes for your website header and app splash screen.
- Promotion cards: square or vertical layouts with clear negative space for text overlays like weekend specials or new-customer offers.
- Neighborhood graphics: stylized maps or skylines that reference service areas without implying coverage you do not have.
- Driver and team content: illustrations or photographs of courier uniforms, delivery bags, and dispatch screens, kept professional and free of consumption.
- Seasonal headers: reusable backgrounds for holidays and local events, with palette instructions that match your current brand refresh.
A simple template you can adapt
Use a fixed structure so results stay comparable across sessions:
Subject: [specific object or person] | Setting: [location, time of day] | Lighting: [light source, quality] | Lens and framing: [focal length, camera angle, negative space location] | Palette: [three hex codes] | Style: [photographic, flat illustration, or similar] | Exclusions: [your standing list]
Save the version that works, note the date and generator version, and retire prompts that stop producing usable output after an update.
Keep a human in the review loop
Generated images are drafts. Someone on your team should review every asset for accuracy, brand consistency, and compliance before it is published. Check small details that generators often get wrong: text on signs, the number of fingers on a courier, the shape of a vehicle door, and any letters that look like real brands. A two-minute review step prevents most embarrassing mistakes.
Log approvals in a simple spreadsheet with the prompt, the output file name, the reviewer, and the channel where it was used. If a regulator, platform, or customer ever asks questions, you will have a clear record.
A final checklist
- Prompt follows your standard template and includes explicit exclusions.
- Output contains no consumption, no minors-appealing styling, and no health claims.
- Any packaging or logo is replaced with approved assets added manually.
- Location and service-area imagery matches your current licensed coverage.
- A named reviewer has signed off, and the approval is logged.
- Local advertising rules and platform policies were checked for this specific channel.
Used carefully, AI prompts can help a small cannabis delivery team produce consistent, on-brand visuals without a large design budget. The businesses that get the most value treat prompts as reusable operating assets, with the same rigor they apply to menus, inventory records, and driver safety procedures.
Leave a Reply