How To Design Prompts That Elicit Creative And Diverse Responses From Generative Models

Generative models are artificial intelligence systems that can produce novel and realistic content, such as text, images, audio, or video, based on some input or prompt. For example, a generative model can write a story given a title, draw a picture given a description, or compose a song given a genre.

However, not all prompts are equally effective in stimulating the creativity and diversity of generative models. Some prompts may result in bland, repetitive, or irrelevant outputs, while others may trigger more original, varied, and engaging outputs. How can we design prompts that elicit creative and diverse responses from generative models?

There are several techniques and principles that can help us craft better prompts for generative models. Here are some of them:

• Use open-ended questions. Open-ended questions are questions that do not have a single or simple answer, but rather invite multiple or complex answers. For example, instead of asking “What is the capital of France?”, which has a straightforward answer (Paris), we can ask “What are some interesting facts about Paris?”, which can generate many different and interesting answers. Open-ended questions can encourage generative models to explore more possibilities and generate more diverse outputs.
• Provide examples. Examples are instances or samples of the desired output or content. For example, if we want a generative model to write a poem, we can provide an example of a poem that we like or that matches our criteria. Examples can help generative models to understand the format, style, tone, or theme of the output we want, and also inspire them to create similar or related outputs.
• Add constraints. Constraints are limitations or rules that restrict the output or content in some way. For example, if we want a generative model to write a story, we can add constraints such as the genre, the setting, the characters, the plot points, or the word limit. Constraints can help generative models to focus on specific aspects or elements of the output we want, and also challenge them to generate outputs that satisfy the constraints while being creative and diverse.

These are some of the techniques and principles that can help us design prompts that elicit creative and diverse responses from generative models. Of course, there are many other factors that can affect the quality and quantity of the outputs generated by generative models, such as the data they are trained on, the algorithms they use, or the parameters they have. However, by applying these techniques and principles to our prompts, we can improve our chances of getting more creative and diverse outputs from generative models.