Multimodal Prompt Engineering with Google Gemini and OpenAI Chat-GPT4 Video

Deepak Bhaskaran
4 min readJan 7, 2024

Are you interested in becoming a world-class prompt engineer specialized in cutting-edge multimodal models? If so, you’ve come to the right place! In this article, I will share five best practices that will supercharge your prompts and turn you into a prompt engineering master.

## Understanding Generative AI and Prompt Engineering

Before we dive into the best practices, let’s start with the basics. Generative AI is a technology powered by machine learning models that can create new content in various forms, such as text, images, audio, and video. Prompt engineering is the art of using natural language to command a large language model or AI language. With the right prompt, you can instruct the AI model to summarize, generate, and review content.

## The Power of Multimodal Models

One of the most exciting advancements in prompt engineering is the development of multimodal models. Traditional multimodal models stitch together text-only, vision-only, and audio-only models in separate stages. However, Google’s Gemini is the world’s first native multimodal model capable of understanding and processing text, audio, code, images, and video inputs simultaneously. This means that a multimodal model can interact with multiple media types just like humans do in the real world, making it incredibly versatile and powerful.

## Best Practices for Prompt Engineering in Multimodal Models

Now that we understand the basics and the power of multimodal models, let’s explore five best practices that will elevate your prompt engineering skills:

### 1. Be Specific in Your Instructions

To ensure accurate and desired outputs, it is crucial to craft clear and concise instructions that leave minimal room for misinterpretation. If you have a specific output in mind, include that requirement in the prompt. For example, instead of asking a model to describe an airport board, you can instruct it to “Parse the time and city from the airport board shown in this image into a list.” By being specific, you guide the model to generate the desired response.

### 2. Add Examples to Illustrate

Multimodal models can benefit from few-shot learning, which involves providing examples to help the model understand the desired output. Adding realistic few-shot examples to your prompt helps the model identify patterns and apply the relationship between given images and responses to new examples. For instance, if you want the model to describe an image, provide a few-shot example that demonstrates the desired output, steering the model in the right direction.

### 3. Break Down Complex Tasks

For complex tasks that require visual understanding and reasoning, it is beneficial to split the task into smaller, more manageable steps. By guiding the model through a step-by-step process, you can enhance its ability to generate accurate and coherent responses. For example, if you want the model to perform a complex task involving image analysis, break it down into smaller goals and list the key steps you want the model to take. This approach helps the model comprehend the task and produce the desired output effectively.

### 4. Specify the Output Format

Sometimes, the downstream tasks that follow model output require a specific format. By specifying the desired output format in your prompt, you ensure that the model generates responses in the format you need. Whether it’s markdown, JSON, HTML, or any other format, prompt engineering allows you to instruct the model accordingly. This practice saves time and effort in converting or manipulating the output to match the desired format.

### 5. Image First for Single Image Prompts

When working with single image prompts, it is generally more effective to place the image before the text prompt. While certain multimodal models can handle image and text inputs in any order, for prompts containing a single image, placing the image or video before the text prompt often results in better performance. However, for prompts where images need to be highly interleaved with text for coherence, use the order that feels most natural.

## Real-World Examples of Powerful Prompt Engineering

Let’s walk through some real-world examples to see these best practices in action:

**Tip 1: Be Specific**

Instead of asking a model to describe an airport board, provide a specific instruction such as “Parse the time and city from the airport board shown in this image into a list.” By being specific, you guide the model to generate a more accurate and desired response.

**Tip 2: Add Examples to Illustrate**

To steer the model in the desired direction, provide few-shot examples that demonstrate the output you expect. These examples help the model understand patterns and apply them to new examples. By adding few-shot examples, you can shape the model’s response according to your requirements.

**Tip 3: Break it Down Step by Step**

For complex tasks involving visual understanding and reasoning, breaking them down into smaller, manageable steps helps the model accurately comprehend and generate responses. By listing the key steps you want the model to take, you guide it through the process and achieve better results.

**Tip 4: Specify the Output Format**

If the downstream task requires a specific output format, instructing the model to generate responses in that format saves time and simplifies the integration with other systems. By considering the desired output format in your prompt engineering, you ensure compatibility and ease of use.

**Tip 5: Image First for Single Image Prompts**

When working with single image prompts, consider placing the image or video before the text prompt. This order often leads to better results, as the model can process visual information before generating the text response.

## Conclusion

Prompt engineering is a fascinating field that empowers us to leverage the capabilities of generative AI models, especially in the realm of multimodal models. By following these best practices — being specific, adding examples, breaking down complex tasks, specifying the output format, and considering prompt order — you can enhance your prompt engineering skills and achieve impressive results. As you continue on your generative AI journey, remember that prompt engineering is both an art and a science, with endless possibilities for creativity and innovation.

Make sure to subscribe to our channel for more informative videos on AI applications and connect with me on my socials. I’m here to help you navigate and excel in your generative AI journey.