Feature
Generating images with artificial intelligence (AI) image generators, such as OpenAI’s DALL-E, is simple but difficult to master. Craft image generation requires both knowledge of AI model's mechanics and goal clarity. In many ways, image generation is akin to a game of artistic telephone. In such a game, each image generation prompt is similar to that of a client and designer, wherein the human expects the AI image generator to deliver excellent visual art while using a few vague phrases and directions.
If you’re looking to improve your prompt approach, consider leaning into the mindset of AI interactions as a role-playing simulation. Here, we explore how you can enhance your image generation by adopting the roles of a design client, design director and patron of the arts.
Role Playing is a Universal Part of GenAI Chatbot Usage
Asking a generative AI chatbot to play a role is a foundational technique of prompt engieering. Humans can play roles also. An article in Nature argues that AI interactions are best viewed as simulations. This perspective suggests that treating AI prompts as role-playing scenarios can activate character archetypes within the training data, leading to more effective outputs.
If you’re looking for copywriting, communicate as an opinionated author, and if you’re looking for a good image, speak as a patron of the arts. To inform the specific requests of your role plays, review prompting guides by Zapier, Microsoft and Foundation. Use Google, YouTube and, increasingly, the large language models (LLMs) themselves, such as Perplexity, for more insight on prompt techniques — even if you’ve been generating images for a while and think you know what you want.
Clients Who Know What They Want Get What They Want
If you’ve ever been in client services you know that the worst part of the engagement is when clients don’t know what they want. Whether internal or external, poor communication and goals can mean weeks of headache and evenings of unnecessary “fire drills.” Input clarity is key for quality output.
Chatting with ChatGPT about creating an image isn’t entirely the same as client interactions, but the principles are similar. Confusing requests may trigger an erratic “fire drill” for the chatbot. Asking an AI to “make it more” is a fun but ineffective way of generating consistent outputs. Vague prompts are great for brainstorming, but specific, concise and detail-aware prompts are the most effective approach.
So even if it’s a simple request, take some time to consider the specifics of your image's desired characteristics. Pulling in clear requirements like your brand kit and dimensions are other useful baselines. Organization and a balance of detail and brevity are the soul of prompting.
If things aren’t working, close the chat and start a new engagement. But if things are going well, consider adding more personality, such as a design director.
Play the Role of a Design Director
A design director is rarely the individual contributor editing images or mock-ups. Like the process of AI prompting, design directors are systems creatives focused on narrative and vibe. The best design directors will go beyond notoriously vague instructions, like “pls fix,” and instead provide a clear frame and principles. Two relevant tactics in this design director role include:
Know Your Whys
At the root of design is knowing the context. Who is this for, and why is it needed? Asking why five times is a general best practice — the five whys approach. Consider an example of a dog logo:
- I want a logo with a dog.
- Why? The dog logo is for a high-end dog-walking service.
- Why? To establish a recognizable and trustworthy brand in the upscale pet care market.
- Why? We aim to convey professionalism and reliability to discerning pet owners.
- Why? To attract and retain clients who seek premium dog-walking services.
- Why? We want to use the logo in all our marketing materials, from business cards to social media profiles.
Share all of those whys to inform the image output. Each why provides deeper insight, helps ensure that your intentions as a client are aligned with your needs and most importantly, provides a more developed context for the image generator prompt.
Treat Generated Images as Prototypes and Temporary Fillers
AI images often have a lot of background noise and issues, which means that production ready may be more trouble than it’s worth. Editing AI images within an image generator UI is still early and often lacks the specificity of manual photo editing. Adopt a prototype-oriented approach for AI images. For instance, use AI-generated visuals as temporary placeholders in drafts, which can later be refined or replaced with higher-quality images.
If you want to continue leaning into the role play, then a patron commissioning a piece is next.
Play the Role of a Patron of the Arts
When you’re working with an image generator you are, in effect, commissioning a piece of art. Like the role of a design director, you have a particular vision for what you’d like to have happen but often with a more technical perspective. Or,at least effective patrons do. They know artistic styles, design elements and other aesthetic preferences — they’re like art critics, knowing generally what they like and what they don’t like. A good place to begin is knowing the underlying prompts of inspirational image generations.
Know Your Inspirations
To start, you can look at the key aesthetic trigger words for example generative AI images at Midjourney’s Showcase. Each image’s corresponding prompt makes it easy to understand some principles of art styles, composition and other technical aspects. It’s also a great way to develop one’s sense of style and aesthetic. Similarly, you can do the same for image generator prompts — check out the image alternative text to investigate how your request was translated. Asking for a specific style, given your prompt, can refine your aesthetic senses more.
Encourage Creativity Beyond the Scope
AI images can be highly effective ways to generate surprisingly creative images. Ironically, intentionally vague prompts are an effective trigger for surprising outputs. After all, if you, as the client, don’t have a mental image of what a “super hip dog living in 2584” looks like, or have a request, like “this is completely wrong, I want the opposite,” then outputs will definitionally be creative. Reviewing the underlying prompts for relevant image generations can also provide refined approaches.
Final Thoughts
Whether you seek creativity, like a “hip dog,” or specificity, like “a dog logo for a Pomeranian-oriented dog walking business,” effectively using image generators boils down to time management and strategic prompting. Prompt opportunities may be infinite, but time and API credits are not. By adopting various role-playing personas, you can streamline your image generation process and achieve better results.
