Feb 26, 20232 min read

Optimizing Language Models: See How Prompt Engineering can Improve NLP Performance

Optimizing Language Models: See How Prompt Engineering can Improve NLP Performance with this Real-World Example #NLP #LanguageModels #PromptEngineering

As covered in my "What is Prompt Engineering" the first 4 steps in designing a prompt are about making sure you have the right materials and data to get the best value out of the prompts you are going to run.

Having a sample or imagery that is say 80% of what you are shooting for to use as a base is an example of being better prepared with your data. However, if you don't have these assets, then creating a solid script can be very effective to fill that gap.

Here is an #Midjourney example of a prompt engineering script for an image needed for a marketing campaign for a Cat Toy company. Their brand and colors are Green, White, and Black. In the creative brief, I start shaping the script from the context, so that I can get to a version quickly that could be exported into Adobe Photoshop for final edits.

Here is the script:

Please generate a realistic 4K image of a cat playing with a green ball of yarn on a carpet, with the letter 'D' in the background.  
Here are some guidelines to follow:  
- The image should be realistic and high-quality, with clear details and colors. 
- The cat should be clearly visible and in focus, playing with the green ball of yarn on the carpet. 
- The green ball of yarn should be a prominent feature of the image and should be easy to distinguish from the carpet. 
- The letter 'D' should be visible in the background, but should not detract from the main focus of the image.  

Please use these guidelines to generate an image that meets the specified criteria. Thank you!

If you were going to manually use the /Imagine prompt.

PROMPT: #Midjourney - /Imagine prompt - cat playing with green ball yarn on carpet with letter D, realistic, 4k