Imagen 3

Imagen 3, the latest text-to-image model from Google, is available for public use, for free, through their AI Test Kitchen website. You do need to sign in with your Google account, but those are available for free if you don't have one.

At the time of this review, Google's Gemini AI is still using Imagen 2, an earlier version that produces lower resolution images. Even though Google Gemini will draw you a picture when you ask it to, you're better off using the AI Test Kitchen if you want Imagen 3.

Google's Imagen 3 is similar to its leading competitor, Open AI's DALL-E 3. Both give you good prompt adherence, so you can just type what you want to see in plain English. Like DALL-E 3's free tier, Imagen 3 restricts you to square images, of 1024x1024 resolution (there's no option for widescreen or taller images). Like DALL-E 3, Imagen 3 is heavily censored, so you shouldn't expect it to generate anything that it might categorize as offensive.

After you type a prompt, you are presented with four possible image choices. But, with Imagen 3, each image has a button labelled "Edit Image" in the lower left corner. When you click Edit Image, you're allowed to paint a simple mask around any areas you want changed. You can also type a prompt, saying what you would like to see in the area you just marked. For example, if you wanted a lizard to have bigger eyes, you could paint a mask over the eyes in the image and type the prompt "big blue eyes." Imagen would then regenerate the parts of the image that you masked out and offer your four choices of new eyes to choose from, seamlessly integrated into the original image.

This editing option, called 'inpainting', can make a big difference. For any image you generate, it's likely that there's at least one small area that you wish had come out differently. With inpainting, you don't need to make a whole new image, but can keep the parts you like, and only replace certain areas of the image.

The AI Test Kitchen website also has an interesting interface, in which words in your prompt are replaced by drop-down menus showing possible alternative words and phrases you could try. It also has words suggested at the bottom for styles of image, such as '35mm film', that you can click on to add to your prompt. These can be useful suggestions for improving or editing your prompts, but if you don't like them, you can still edit the prompt conventionally as well.

Google's Imagen 3 is a very solid competitor among free on-line image generators. With good prompt adherence and the option to inpaint whatever areas of the image you want changed, it seems like it's worth trying for anyone who wants to quickly visualize a prompt.