Day 2 of Building with Gemini
Exploring Gemini - Comparisons in Multimodal Contexts
Today we’re going to dig in with some fun Gemini prompts in Google AI Studio that lean into the idea of comparisons, inspired by my everyday life.
Healthy Eater
In this first example, I compare 3 different salads that I’ve made:
Thai beef salad
Grilled caesar salad wedges
Breakfast salad with bacon and avocados
Side note: I love cooking, and I really love salads
You’ll see that Gemini was able to identify various characteristics of each salad, then determine which one was healthiest with reasoning around fat and calorie information it inferred from the photos.
Odd One Out
In this example I wrote a prompt that instructs Gemini to identify which image doesn’t belong, using some of my daughter’s favourite toy animals.
You’ll notice that Gemini is able to first identify what each of the items are, and then describes different characteristics and traits about them so that it can reason about why one of them shouldn’t be included in the set.
Festive Choices
In this example we’ll compare napkins to help us determine which is the best option for a holiday gathering.
Comparison Prompt Template
The Gemini multimodal API is great at being able to compare different images and draw conclusions about aspects of them with reasoning and logic.
In general, prompt engineering is an iterative process. Try out an initial prompt, see where it might be failing, and then go back and improve your prompt based on that. I find the following prompt template a helpful place to start:
Bring on the building!