Many small delivery teams have started experimenting with image generators to produce menu graphics, seasonal banners, and social posts, and quickly discover that the results are inconsistent. Some teams decide the fastest route is to buy midjourney prompts from a marketplace instead of learning every trick through trial and error. Whether you build your own library or purchase one, the underlying question is the same: what separates a prompt that actually works from one that only looks good in a sample post?
Why most prompts fail
Most disappointing prompts share a few traits. They are vague, they describe a mood without describing a scene, and they leave the model to guess about composition, lighting, and style. A prompt like “a cozy winter evening with cannabis products” will return something, but rarely something you can use. The model fills every gap with its own defaults, which means your brand voice and visual identity are lost.
There is also a problem of reproducibility. A prompt that produces one excellent image by accident is not a working prompt. A working prompt produces acceptable results repeatedly, across different subjects and sizes, with only small adjustments. If you cannot predict what a prompt will do, you cannot build a workflow around it.
What a working prompt contains
After reviewing many prompts over several months of use, a consistent structure tends to emerge. Strong prompts usually include the following elements:
- A clear subject. Name the object or scene in concrete terms, such as a glass jar on a wooden tray beside a window in late afternoon light.
- A defined composition. Specify framing, camera distance, and where the subject sits in the image.
- Lighting and color direction. Describe the light source and a limited palette. Two or three colors are easier for a model to hold than ten.
- A style reference that you control. Instead of naming a famous artist, describe the qualities you want: matte textures, clean editorial layout, muted greens.
- Negative constraints. Tell the model what to avoid, such as cluttered backgrounds, text artifacts, or people in the frame if your use case does not call for them.
- Aspect ratio and use case. A banner for a storefront window needs different proportions than a square social tile.
Notice what is missing from that list: magic words. Most of the value in a prompt comes from clarity and restraint, not from stacking adjectives.
Adapting prompts to a regulated product category
A cannabis delivery business in Quebec operates under strict rules about how products can be promoted. Federal cannabis legislation and Quebec provincial rules restrict advertising, limit appeal to young people, and control certain claims. Any image or text generated by AI is still your publication, so the responsibility for compliance stays with you. Treat generated output the same way you would treat work from an outside designer or copywriter.
In practice, that means building prompts that avoid lifestyle imagery suggesting consumption, avoid anything that resembles children or youth-oriented styling, and avoid health or therapeutic claims. A prompt for a delivery promotion might focus on packaging, clean delivery vehicles, the bag or box design, or abstract brand graphics rather than people using products. It is also worth writing a short internal policy listing banned themes so that every team member writes prompts against the same boundaries.
Before any asset goes live, have someone check it against current regulations. Rules change, and an image that was acceptable last year may need revision now. If your business has legal counsel, ask them to review your prompt guidelines once, then keep the approved list as a reference.
Building a testing habit
The fastest way to improve prompts is to test them deliberately. Choose one task, such as a weekend promotion banner, and write three versions of the prompt that change only one variable each. Maybe one changes the lighting, another changes the composition, and a third changes the color palette. Generate several images from each version, then compare them against a simple scorecard.
A scorecard can be as basic as four questions: Does the image match the brief? Is the subject clear at thumbnail size? Does it fit our brand colors? Would we be comfortable publishing it without editing? Record the results in a shared document. Over time, you will build a library of tested prompts with notes about what each one does well and where it fails.
This is also where purchased prompts can be useful. Teams that want to compare ready-made options can look through a curated library of tested image and text prompts at PromptMart’s prompt marketplace, then adapt the most promising ones to their own compliance rules and brand guidelines. Treat any purchased prompt as a starting draft, not a finished asset, and run it through your testing process before relying on it.
Keeping prompts organized
A prompt library becomes valuable only if people can find and trust it. Organize entries by use case rather than by model or date. A useful entry includes the prompt text, the intended format, the model and version it was tested on, a short note on known problems, and the name of the person who approved it. Store everything in one place that the whole team can access.
Version control matters too. Models change, and a prompt that worked well on one release may behave differently on the next. When you notice a drop in quality, add a dated note rather than overwriting the old version. That history makes it much easier to diagnose problems later.
A short workflow for delivery teams
- Write a one-paragraph brief that states the goal, audience, format, and restrictions.
- Draft a prompt using the subject, composition, lighting, style, constraint, and format structure described above.
- Generate a small batch of outputs and score them against your checklist.
- Revise only one variable at a time and retest.
- Run the final candidate past a compliance reviewer before publication.
- Save the approved version to your shared library with notes.
This workflow is slower than simply typing a request and accepting the first result, but it saves time over the long run. Once a team has a dozen approved prompts for recurring tasks, a new banner or social tile can take minutes instead of hours.
Final thoughts
Prompts that actually work are specific, reproducible, and matched to a clear purpose. For a cannabis delivery business in Quebec, they also need to respect advertising rules and the expectations of the communities you serve. Whether you write every prompt yourself or draw on outside libraries, the discipline is the same: define the goal, test deliberately, document what you learn, and keep a human reviewer in the loop. Done well, AI image and text tools can help a small team look professional without stretching its staff or budget.
Leave a Reply