Image generation is a staple feature of AI chatbots. To start, I tested how well the chatbots could make photorealistic images without errors or distortion. I used the following prompt in Gemini (Nano Banana Pro), ChatGPT (GPT Image 1.5), and Copilot: “Generate me an image of a cozy suburban home with an open floor plan. I want to see a nice living space with a dining room, kitchen, and living room. Nothing too fancy.” Below are the results from Gemini (first slide), ChatGPT (second slide), and Copilot (third slide):
Gemini’s image looks excellent. It’s lifelike, richly detailed, and free of major errors or distortions. ChatGPT’s image is similarly high-quality, but it’s slightly less detailed than Gemini’s. Copilot’s image doesn’t have any serious problems, but the apartment it depicts looks more like a stock photo than one where actual humans live.
Next, I tested the chatbots’ ability to generate complex illustrations: “Generate me a six-panel comic image of a high fantasy kingdom where wizards have replaced spellcasting with PowerPoint presentations. Make sure there’s a major twist in the final panel.” Here are the results from Gemini (first slide), ChatGPT (second slide), and Copilot (third slide):
Gemini’s comic is, again, the best of the three. Not only is it illustrated well and packed with detail, but its text generation is flawless and borderline amusing. ChatGPT’s comic doesn’t have any major issues, and it tells a reasonably coherent story, too. Copilot’s comic lacks a twist, has a nonsensical story, and uses a simplistic art style.
Get Our Best Stories!
All the Latest Tech, Tested by Our Experts
Thanks for signing up!
Your subscription has been confirmed. Keep an eye on your inbox!
My final test was to generate a technical diagram with the following prompt: “I’ve got an Ethernet switch, a modem, and a router. I want to connect my PC, PlayStation, smart light hub, and smart TV via Ethernet. I want the modem to connect to the router, and then the router to connect to the switch. Then, I want the switch to connect to the devices. Generate me a diagram showing this.” You can find Gemini’s (first slide), ChatGPT’s (second slide), and Copilot’s (third slide) results below:
Gemini’s and ChatGPT’s diagrams are both high quality, but they also have some minor issues. For example, Gemini’s PC connection arrow is backwards, and ChatGPT has a strange connector to the right of its Ethernet switch. Copilot’s diagram is passable, but its art style is simpler. It’s also missing a connection between its modem and router.
Image Editing: Gemini Just Barely Beats ChatGPT
Beyond image generation, Gemini offers AI image editing. Simply send Gemini a photo and describe the edits you want. AI-edited images often look seriously distorted or obviously AI-generated, but Gemini generally performs better than other chatbots in this regard.
I asked Gemini (Nano Banana Pro), ChatGPT (GPT Image 1.5), and Copilot to remove my hand from an image I provided in testing. You can see the results from Gemini (first slide), ChatGPT (second slide), and Copilot (third slide) below:
No chatbot handled my prompt perfectly. Gemini stood out for its ability to match the original image’s aspect ratio, which ChatGPT and Copilot didn’t, and its image had the highest resolution of the bunch. However, Gemini’s image is somewhat grainy and washed out. ChatGPT’s image, on the other hand, was almost too vibrant, and tessellation across its finer details was distracting. Copilot’s image matches the color of the original well, but it’s fairly grainy.
Editing goes beyond making simple alterations to images. AI chatbots can also blend images. To test this feature, I took a picture of my kitchen, and then provided images of a table and some chairs from Amazon. Next, I prompted the chatbots to add the chairs and table into the photo of my kitchen, making it look as realistic as possible. Gemini’s (first slide), ChatGPT’s (second slide), and Copilot’s (third slide) edits are below:
Once again, no chatbot did a perfect job. Gemini’s image matches the original’s aspect ratio, which no other chatbot managed, and its lighting feels the most realistic of the three. However, Gemini also distorted the background and added a window above the sink for no apparent reason. ChatGPT did a serviceable job, but it also introduced distortion. The lighting of its table and chairs is slightly uncanny, too. Copilot’s image shows the most distortion of the three, resulting in overall poor image quality.
Unlike ChatGPT, Gemini doesn’t integrate directly with Adobe Express and Photoshop. Still, you can prompt Gemini to adjust the same aspects (brightness, contrast, saturation, and more) of an image with just as good (or better) results. Either way, chatbots aren’t a viable replacement for photo editing software.