How to Write an AI Prompt

TL;DR: By adding specific details, leveraging keywords (such as Three.js), using image references, and iterating on your prompts in Google AI Studio, you can build applications that more accurately match your expectations. ## Prompt Specificity Determines Output Quality A well-crafted prompt often determines whether the result is "this" or "that." The Vibe Coding feature in Google AI Studio allows users to build applications through simple descriptions, but many people overlook a key point on their first attempt: you can significantly guide the model’s behavior. If you input an overly open-ended prompt, such as "make me a racing game," the result is usually quite generic. To get output closer to your expectations, you need to add specific details. Clearly specifying requirements using a bulleted list can significantly improve results. For example, instead of just asking for a "racing game," be explicit: * Include a small blue car. * The track must be circular, allowing for full laps, with a finish line. * Include collectible speed boost power-ups. By explicitly defining these details, the model’s output will align more closely with your vision. ## Leverage Keywords to Support Specific Requirements Understanding which keywords trigger the model to build products in the best possible way is highly valuable. Take creating a 3D video game as an example: using the keyword **"Three.js"** in your prompt is crucial. Three.js is a popular 3D graphics library that the model handles exceptionally well. If you’re unsure which technologies to use, simply ask the model. For instance, you might ask, "What are the best web technologies for building 3D games?" The model will recommend tools like Three.js. With a more specific prompt, you’ll see results featuring a blue car, collectible speed boosts, and a circular track style that meets your requirements—getting very close to your intended outcome. ## Image References and Iterative Refinement In addition to text descriptions, image references can play a significant role. Uploading a screenshot alongside your prompt can visually guide the model, helping it render the scene as you envision it. You might phrase it like this: "Look at this screenshot. There’s a detail here that I particularly like; please make sure to add it to our app." Furthermore, don’t settle for the output from your first prompt. The core of Vibe Coding lies in performing a "vibe check" on the product you’re building. If you find during playtesting that the car controls feel slightly off, or if you want the speed boosts to have a stronger effect, go back to the prompt box and iterate on those specific details. For example, you might provide feedback like: "I love the speed boost, but I’d like it to be faster or last longer." By interacting back and forth with the model and continuously adjusting details, you can refine the game feel until it perfectly matches your vision. ## Explore the AI Studio App Gallery Google AI Studio also features an App Gallery filled with templates and examples. Users can find inspiration here and explore the potential of the model. You can add existing apps, Remix (remix/modify) them, and create your own unique versions. This is an excellent place to explore AI capabilities and gain creative inspiration. By increasing prompt specificity, leveraging the right technical keywords, using image references, and iterating continuously, you can achieve the ideal output for your projects.