6 Tips to Help You Get Started with AI Image Generation

Updated March 2026.

AI image generation has moved from novelty to professional tool in a short period of time. DALL-E from OpenAI was one of the first widely accessible tools. Since then, Midjourney and Adobe Firefly have joined the field, each with different strengths, interfaces, and licensing terms. All of them work on the same basic principle: you describe what you want, and the model generates an image.

The quality of what you get depends almost entirely on the quality of what you ask for. These are not point-and-click tools. They reward designers who can articulate what they are looking for — which, if you think about it, is a skill at the heart of design itself.

Before we get into the practical tips, it is worth reading our post on using DALL-E ethically, which covers copyright, disclosure, and appropriate use. The tips below focus on getting better output; the ethical post covers what you should and should not do with it.

When you are starting out, the most useful thing to understand is that the AI is only as good as the information it receives. Vague prompts produce vague images. Specific prompts produce useful ones.

Start simple and concrete. “A red apple on a white plate” will give you something coherent. “A concept” will give you something generic. The tool cannot read your intentions — it reads your words.

Simple, specific prompts also help you build a mental model of how the tool interprets language. You learn what kinds of descriptions produce what kinds of results. That understanding compounds quickly once you start paying attention to it.

The more specific your description, the more likely the result is to match what you are imagining. That means describing not just the subject, but the mood, the lighting, the colour palette, the medium, and the context.

Compare these two prompts:

• “A flower”
• “A bright pink peony, overhead view, soft natural light, minimal white background, editorial style”

The second prompt gives the model much more to work with. You are essentially writing an art direction brief — which is exactly the skill that makes experienced designers better at using these tools than non-designers.

Useful descriptors to layer in: lighting conditions (golden hour, overcast, studio), rendering style (photorealistic, watercolour, flat illustration, engraving), mood or atmosphere, camera angle or distance, and any colour constraints.

One of the most powerful aspects of AI image tools is the speed of iteration. What would take hours to sketch and redraw can be explored in minutes. The constraint is your imagination, not your drawing time.

Start by brainstorming combinations of objects, settings, and styles that interest you. Mix things that do not obviously belong together. The results are often more useful than the obvious interpretation — they can point you toward something unexpected that genuinely works for your brief.

The bears in the images on this page were generated through exactly this kind of iterative process: starting with a simple concept and adjusting the prompt with each generation until the result was closer to what we had in mind. Each generation teaches you something about the tool.

Metaphors are not just literary devices — they are useful prompt tools. Describing something in terms of something else can produce results that a literal description would not.

Try prompts like: “A city at night, like a circuit board seen from above” or “A wine label that feels like a faded textile from a 1920s French market.” The figurative language gives the model a richer set of associations to draw from.

Be specific within the metaphor. “Like a painting” is too broad. “Like a Hopper oil painting, warm interior light, isolated figure” gives the model a much clearer visual reference to work with.

Designers think about composition — AI tools respond well to compositional language. Tell the model what perspective you want: bird’s-eye view, close-up detail, wide establishing shot, portrait orientation. These are the same decisions you make when art directing a photography shoot or laying out a spread.

Composition language that tends to work well includes: “overhead flat lay”, “wide angle establishing shot”, “tight crop on hands”, “symmetrical composition”, “rule of thirds”, “negative space dominant”.

Thinking in compositional terms also helps you connect AI-generated assets back to your design layouts. An image generated without a compositional intent is harder to integrate into a grid than one that was guided toward the space you need to fill.

This is not a legal guide, but it is a necessary conversation. The copyright status of AI-generated images is genuinely unsettled, and it differs by tool and jurisdiction.

DALL-E and OpenAI: Current OpenAI terms grant users rights to use generated images commercially, including rights to reproduce, sell, and merchandise. However, you cannot claim copyright over the output in most jurisdictions, because copyright requires human authorship.

Midjourney: The licensing terms have changed over time. On paid plans, users retain rights to commercial use. On free plans, images are licensed under Creative Commons (non-commercial). Check the current terms before using Midjourney images in client work.

Adobe Firefly: Adobe built Firefly specifically to be commercially safe — it is trained on licensed Adobe Stock imagery and public domain content. For professional use where IP indemnification matters, Firefly is currently the most defensible option.

The practical advice: be transparent with clients about the provenance of images. Do not represent AI-generated images as original photography or illustration. Keep records of prompts and generation dates. As this space evolves, those records will matter.

These tools reward time spent with them. The designers who get the most out of AI image generation are not necessarily the most technically skilled — they are the ones who experiment most, who iterate without attachment, and who treat each generated image as a starting point rather than a finished asset.

Start with prompts for personal projects, not client work. Build your mental model of how the tool thinks. Then, when you bring it into professional projects, you will know what you are asking for and how to get closer to it.

For the ethical framework that should underpin all of this, read How to use DALL-E ethically in your graphic design work. And if you want to build the design foundations that make AI tools genuinely useful, explore our graphic design courses — the thinking comes first.