AI tools
February 9, 2024

Bard Now Supports Image Generation

Google finally added image generation support to Bard.

Jim Clyde Monge
by 
Jim Clyde Monge

After years of holding off, Google has finally added image generation to its AI chatbot Bard, catching up on a feature rival ChatGPT Plus has offered for months. The difference? Bard is still completely free, while ChatGPT Plus costs $20 per month.

Create images with Bard

What: You can create captivating images for work, play, and anything in between with Bard. It’s easy to use — simply enter a few words to bring your imagination to life, starting with English prompts. Click ‘Generate more’ for more options and download the ones you like. Learn moreOpens in a new window.

Why: Sometimes you might have a picture in mind and now you can easily bring that to life. Create totally unique images in seconds without having to learn complex software.

How Image Generation Works

Head over to https://bard.google.com/ and start generating images by describing them in the text prompt. Here’s an example:

Prompt: an close up image of a cute dog

Google bard an close up image of a cute dog
Image by Jim Clyde Monge

Awesome. Now, what’s great about it is that you can make further modifications by simply asking the AI. For instance, I wanted to turn the image of the dog into Disney Pixar style.

Prompt: turn the image into Pixar 3D style

Google bard image of a dog turn the image into Pixar 3D style
Image by Jim Clyde Monge

How cool is that? Another trick is to upload an existing image and ask Bard to make variations of it.

Google bard generate variation of this logo image
Image by Jim Clyde Monge

Sure, it can generate image variations, but often not what you’d expect.

It Can’t Generate Human Faces

Before you start thinking about generating realistic human faces, Bard does not allow it because it goes against Google’s AI Principles.

Here’s an attempt to generate a human portrait:

Prompt: an image of a girl model, close up portrait photo, soft tones, anime style

Response: I can’t generate the content you’re requesting because it depicts a real person. It is against Google’s AI Principles to return content that could be used to identify or harm an individual.

However, I can generate an image of an anime-style girl with soft tones. Would you like me to do that?

Google bard. Prompt: an image of a girl model, close up portrait photo, soft tones, anime style Response: I can’t generate the content you’re requesting because it depicts a real person. It is against Google’s AI Principles to return content that could be used to identify or harm an individual. However, I can generate an image of an anime-style girl with soft tones. Would you like me to do that?
Image by Jim Clyde Monge

Bard suggests prompt modifications that do not go against their privacy rules.

Similarly, you can’t upload an image of a person and ask Bard to make modifications or variations of it.

Google bard cannot help generate variations of a human portrait due to AI principle issue
Image by Jim Clyde Monge

Let’s Talk About Image Quality

To be honest, I was super excited when the image generation feature was rolled out today. My expectations were high since Google announced Imagen 2.0 with incredibly high-quality results.

But I was wrong…

The very first image I generated with Bard was far from what I expected. Here’s the comparison between an image presented in Imagen 2.0’s announcement and an image I generated with Bard today.

Imagen 2.0 vs Google Bard image generation
Image by Jim Clyde Monge

Seriously Google?

It was clear in an X post that Imagen 2 powers the image generation capability of Bard.

But what is going on with the low-quality results I am getting?

Try It With ImageFX

The new image generation capability is not just available in Bard. Google also rolled out this feature on its AI Test Kitchen, and they are calling it ImageFX.

To try, head over to the ImageFX website and login with your Google account.

The new image generation capability is not just available in Bard. Google also rolled out this feature on its AI Test Kitchen, and they are calling it ImageFX.
Image by Jim Clyde Monge

Unfortunately, though, this is only available in the US right now.

Final Thoughts

Overall, I am happy to see that Google has finally rolled out the long-awaited image generation feature in Bard. Unfortunately, the quality of the images I generated was a bit far from what I initially expected.

It’s understandable that there will be some limitations in this first iteration. I’m sure it will improve rapidly with more training data and model tweaks.

And, oh, guess what? The sample photo of a realistic human portrait on the Imagen 2.0 announcement page was removed and replaced with an image of a dog.

I hope Google refines their image model to generate higher-quality, more creative results comparable to Midjourney and Dall-E 3 over time. Still, this is an exciting new capability, despite some initial shortcomings.