It’s no secret that AI is increasingly used to create art and images. But how do the leading AI platforms — Google Gemini, OpenAI's ChatGPT, and Meta AI — fare when prompted to generate a delicious three-course meal?
To find out I put them to the test with a series of prompts. The goal was to challenge them to design a creative and appetizing appetizer, entrée, and dessert.
I explored each AI model’s capability based on the context of the food imagery and compared how each platform interprets prompts for each particular course. From appetizers to desserts, let’s see which AI serves up the most appetizing virtual meal.
Creating the prompts
To keep the comparison fair, I began by telling each AI model that I wanted to create a cohesive meal with three courses. From there, I crafted three specific prompts for each course of the meal: appetizer, main course, and dessert.
The challenge for each AI model was to generate visually appealing and realistic food images, while also testing their ability to capture texture, composition, and creativity in a single frame.
Each prompt was chosen to encourage the AI model to stay creative while also maintaining a degree of realism that one might expect from a gourmet menu. My goal was to see how well each AI could deliver not just a beautiful image, but also something that looked mouthwatering and delicious.
Workflow
For this comparison, I used each platform’s image-generation features with their default settings. I did not use any premium models although Google Gemini and ChatGPT both offer them. Each AI image generator was tested through conversational prompts.
The images were assessed based on realism, and how well they creatively captured the essence of a gourmet dining experience. Let’s dive in and see how well the AI platforms performed.
1. Appetizer
Prompt: I’m creating a gourmet menu with three courses. Please create an image of the appetizer.
Gemini: If someone offered me this appetizer, I might politely say that I was saving my appetite for the main course. This simply isn’t appealing. It’s hard to tell if that is hummus in the middle or if the veggies are wilted or roasted. Overall, this gives a centerpiece vibe, not something worth sinking my teeth into. Even the presentation falls flat; like a charcuterie board put together by a toddler.
Meta AI: The shrimp cocktail looked amazing! The salad had strong color contrast, texture, and the image was overall very appealing. With it’s fine details and lifelike plating, it looked like a photo in a magazine.
ChatGPT: This caprese salad was visually appealing but had an artistic, slightly surreal quality. The colors were bold, and the textures were emphasized, though the realism was must better compared to Google’s appetizer. The presentation was well done, fancy but not overly so. The bread added appeal and dimension, which I appreciated after Google’s lack of depth and style.
Winner: Meta AI: The real contest here was between ChatGPT and Meta AI as they both offered appetizers with creative appeal, realism, and attention to detail. But it was Meta AI that delivered a solid first course that captured next-level intricacy.
2. Main course
Prompt: I’m creating a gourmet menu with three courses. Thank you for the appetizer. Please create an image of the main course.
Gemini: The filet mignon atop mashed potatoes looked restaurant-ready, delicately accented with a luscious sauce pattering the plate. This dish was beautifully plated and the texture stands out with detail you can almost taste.
Meta AI: The main course had a stylized and mouthwatering look, with a clearly recognizable dish. Everything gave off mouthwatering appeal right down to the delicate details of the seasoning. The image was not only pleasant but had intense photorealism that made this meal appear extremely delicious.
ChatGPT: This entrée was similar to Gemini’s main course but the asparagus and carrots give off an exaggerated and unappetizing image. The sauce or gravy surrounding the dish was an odd color and texture that leaned more towards art rather than realism. The texture here was not as shar or detailed as Google’s or Meta’s AI model generation.
Winner: Meta AI: Meta AI’s image once again came out on top for realism and presentation. ChatGPT and Gemini both offered creative interpretations, but the lack of fine detail in Gemini’s version made it less convincing and ChatGPT’s dish just didn’t give off the enticement as much as Meta AI.
3. Dessert
Prompt: I’m creating a gourmet menu with three courses. Thank you for the appetizer and main dish. Please create an image of the dessert.
Gemini: The perfectly round strawberry layer cake was gorgeously presented, with delicate sauce drizzled around the plate. The layers added a pop of color that contrasted beautifully with background and the strawberries atop the cake were textured perfectly with extraordinary detail.
ChatGPT: The molten lava cake had a whimsical, almost cartoonish quality, with exaggerated details like the flowing chocolate and a large scoop of ice cream. While visually fun, it didn’t have the realism of Google’s image.
Meta AI: The ice cream sundae gave off a 1950’s American diner vibe that fell flat—a surprise after two rounds of appetizing fare. The scoops of ice cream atop the banana almost look too perfect and a final dollop of whipped cream was missing on the last scoop. The chocolate syrup drizzle lacked the fine detail that would make the dessert truly stand out even if it fit the menu.
Winner: Gemini: Simplistic and understated, Gemini Imagen delivered a rendering of a gourmet dessert with appeal. ChatGPT leaned too far into its artistic side with a playful interpretation and Meta AI’s version was attractive but lacked cohesion and refinement.
Overall winner: Meta AI
After comparing all three AIs, it’s clear that Meta AI consistently delivered the most realistic, detailed, and appetizing images. Despite choking on the last round, Meta AI generated the most photorealistic food imagery.
ChatGPT produced more creative, slightly surreal images that leaned toward artistic interpretation, which could appeal to users seeking more abstract representations.
Gemini, while strong in overall design, didn’t quite match the level of realism or detail provided by Meta AI.
When it comes to generating the most realistic images, Meta AI stands out, but for creativity and artistic flair, ChatGPT is a solid contender. Gemini offers good results but can fall short when it comes to intricate detail.
More from Tom's Guide
- Meta Movie Gen looks incredible — 5 impressive examples
- Apple is '2 years behind' on AI but here's why I think it will still end up ahead
- I asked ChatGPT to create images based on what it knows about me — here's how it went
