If 2023 was the year generative AI went mainstream, 2024 is the year AI will get real. Experts predict the technology could add up to $4.4 trillion annually to the global economy. But it will only reach that potential if users can leverage the tool to improve productivity and produce better results.
Enter: prompt engineering — the practice of writing optimized inputs (prompts) to produce better generative AI outputs.
We recently sat down with Barak Leon, General Manager and Co-founder of FinalTouch, a new venture at Cloudinary, to learn how prompt engineering can help companies create better outputs, improve product development, and realize the full value of generative AI.
Prompt engineering is the process of crafting effective and precise instructions for AI models to generate desired outputs, from text and data to images and videos. “It’s like whispering codes and instructions to the heart of a machine,” says Leon.
A well-written prompt can make all the difference in the final result. For example, a simple prompt like “Write a 150-word social media post for a clothing brand” might produce a generic, bland piece of content. In contrast, a more specific, detailed prompt could deliver the branded, engaging social media post audiences crave.
Here’s a more contextual, unambiguous prompt example:
“Write a 150-word social media post for a clothing brand focused on comfortable basics. Gear the post toward fashion-forward, on-trend women ages 18-30. Make it light-hearted and fun but not cheesy. Avoid using emojis and exclamation points.”
Obviously, writing a prompt with this level of precision takes more thought and understanding of both the audience and the tool. This is why Leon considers prompt engineering a true art of language, one that transforms complex algorithms into approachable tools.
Leon emphasizes that prompt engineering isn’t always completely intuitive, so it’s important to learn best practices that help hone the skill.
Here are his top four tips for crafting precise cues that will generate more specific, accurate outputs:
The most effective prompts vary by industry and desired output, so refining strategies through trial and error and continual experimentation is important. You may need to try several different prompts before landing on the most accurate, relevant output.
Prompt engineering requires a careful balance between simplicity and complexity. Vague prompts may produce irrelevant outputs, while ultra-specific prompts may confuse AI. Engage with the AI community to learn from more experienced prompt engineers, who can share their wisdom and help you produce desired outputs.
Once you’ve discovered a prompt style that’s working, implement your skills in your products so you can enhance other users’ experiences. Continue to iterate and refine based on continuous feedback.
Fully delve into the AI models you work with. Take time to understand their quirks, strengths, and limits. While this is not a must, it can make the difference between good and great while shortening trial and error cycles.
As we move into the future, humans will increasingly look to AI as a co-pilot, and prompt engineering will play a critical role in fostering that relationship. Still, many companies struggle to master the art of prompt engineering, which is why Cloudinary takes all the guesswork out of the process.
Our AI-powered platform FinalTouch handles the complexity of prompt engineering, allowing users to achieve amazing visual media results without worrying about crafting the perfect input.
Barak adds, “It simply makes no sense to spend a fortune on pro cameras or say that you want your AI photoshoot to “not be deformed” in order to get the results you want. In the near future, technical, unintuitive prompts will be replaced with direct, simple, specific prompts. This is the approach we’ve built into FinalTouch. The idea is to make a tool you can master in two minutes.”
Get in touch today and learn how to transform generic product images into engaging, brand-friendly scenes with the power of AI.
If you found this blog post helpful and want to discuss it in more detail, head over to Cloudinary Community forum and its associated Discord.
