I Ate AI Generated Meals For Two Weeks. Here's What Happened.

None of the food below is real. Each image was generated using the latest Stable Diffusion model prompted with recipes generated by GPT4.

Background and Motivation

Over the past few weeks, my kitchen has been something of a science experiment. I've been cooking dinner for my family using recipes created entirely using our FREE AI Recipe Generator.

I did this because I was tired of cooking the same meals over and over for my family. I wanted something that would give different and unique meal ideas to break up the monotony of our weekly menu.

The process was pretty straightforward: ChatGPT provided the recipe titles, listed out the ingredients, and gave me step-by-step cooking directions. But what's a recipe without a visual, right? To give the recipes some pizzazz, I used Stable Diffusion to generate images for each dish.

How did things turn out?

Overall the meals turned out great. The ingredients and cooking instructions matched up perfectly and the instructions were clear and easy to follow. As for the images, the Stable Diffusion model churned out some remarkable visual representations of the meals.

Sure the images aren't perfect. Close inspection reveals inconsistencies that give away their true origin. But they were close enough to fool my brain into excited anticipation while preparing each meal.

In the time since I first published this article I've gone on to cook dozens of AI generated recipes and built an extensive list of My Favorite AI Generated Recipes along with my experience cooking each dish.

Meal Planning Made Easy

One of the biggest benefits was just how easy it made it to plan our meals for the week. Instead of sifting through countless websites, blogs, and cookbooks to find that perfect dish, I simply relied on the AI to produce varied and creative meal ideas. It not only saved me a considerable amount of time but also introduces me to new flavor combinations I might not have considered otherwise.

Minor Complaints

If there's one complaint it's that a few of the recipes were a bit plain. A sauce here or a few extra ingredients there would have made the meals more exciting and flavorful. But I suppose that's something that could be fixed with better prompting.

Closing Thoughts

This experiment was thought provoking for a number of reasons. As a software developer seeing my app come to life with real data was incredibly motivating.

But it also raised some interesting questions. One of my goals for MealPractice is to showcase actual meals being cooked, with the assumption that those cooked most often are worth recommending.

But does an AI that learns your preferences and generates custom recipes for you make that obsolete? After working with these meals for the last few weeks I don't believe so. I still find a recipe that has been vetted by other users to be more appealing than something that hasn't received any human validation. At least for now.

If you thought this was interesting feel free to say hi on twitter @benbjurstrom. Thanks for reading!