Google Claims Imagen 3 AI Image Generator Surpasses DALL-E 3: Here's How to Test It

Choosing the right AI chatbot can feel overwhelming with so many options out there. Google DeepMind stepped in to help, conducting a thorough comparison of top chatbots. Surprisingly, it was an image generator that caught everyone's eye—Imagen 3.

I recently gave Google's ImageFX AI image generator a whirl, and I must say, the results were stunning.

Google DeepMind's detailed report, released on Wednesday, showcases how Imagen 3 stacked up against its predecessor, Imagen 2, and other leading models like DALL-E 3, Midjourney v6, Stable Diffusion 3 Large, and Stable Diffusion XL 1.0. The study used both human and automatic evaluations to assess performance.

Human evaluations focused on five key areas of text-to-image generation: preference, prompt-image alignment, visual appeal, detailed prompt-image alignment, and numerical reasoning.

In the overall preference category, which gauges user satisfaction with the generated image relative to the input prompt, Imagen 3 significantly outshone its competitors, as illustrated in the image below:

Imagen 3 held its own in the other human evaluation categories and in the automatic evaluations, which also assessed prompt-image alignment and image quality.

The report concluded, "All in all, Imagen 3 clearly leads on prompt–image alignment, especially on detailed prompts and counting abilities; while on visual appeal, Midjourney v6 takes the lead, with Imagen 3 coming in second."

"When considering all the quality aspects, Imagen 3 clearly leads in overall preference, indicating it strikes the best balance of high-quality outputs that respect user intent."

Skeptical? You can test Imagen 3 yourself through ImageFX, a tool available in Google Labs that lets you create images with simple text prompts.

How to Use ImageFX

ImageFX is part of Google Labs, where the company experiments with new ideas and products. Accessing it is easy and free.

Just head over to Google Labs and select ImageFX, or go directly to the ImageFX page. Sign in with your personal Google account, and you're ready to start playing around. Enter a conversational prompt describing what you want to see.

One of the fun features of ImageFX is its "expressive chips." These allow you to explore "adjacent dimensions of your creation and ideas," as Google puts it. Once you type in your prompt, toggles will appear on selected words, suggesting new and amusing ways to tweak it.

Each generation produces four high-quality images for you to enjoy. From my own experience, ImageFX even managed to render hands accurately, which is often a challenging task for image generators.