Google Is Incorporating Its Image-Generating A.I., Imagen 3, into Gemini Advanced

The Five Most Key Takeaways from This Blog Post

Google’s latest version of its image-generating A.I. is Imagen 3. Subscribers to the Gemini Advanced service will have access to it.
This comes after just about half a year of Google laying low from the image-generating lane of the A.I. race after Bard–the old-name version of its generative A.I. platform–struggled to satisfy pretty basic requests, while also generating some offensive output. As a result, for some time users were unable to generate images of people.
Google is also allowing corporate customers to create “Gems”, customizable versions of Gemini that can serve different functions. Access is granted to Gemini Advanced, Business, and Enterprise customers.
Some potential issues may arise from businesses using Imagen 3. For one, the platform has been shown to generate picture-perfect replications of real-world companies’ logos.
Business owners with access to Gemini Advanced may find some uses for the Gems feature, but should expect that Imagen 3 will not be perfect.

What Is Imagen 3?

One of the most difficult of the petty frustrations that comes with trying to keep up with the A.I. race is that there are just so many product names, including rebrands, to keep up with. It can be hard to keep things straight, really.

Imagen 3 is like Google’s version of OpenAI’s DALL-E, in that they are both like Large Language Models (L.L.M.s; these were trained on a wide range of text to generate a wide range of text based on prompts) but for images.

That means you can generate everything from “photorealistic” images that look like camera snapshots of the real world, to photos that look like oil paintings.

Gemini Is Back in Action

Yes, after much controversy, Google is now letting Gemini Advanced users generate images of people again.

Use this at your own discretion, because one of the risks here is that Google’s A.I. can be pretty unpredictable. It remains to be seen just how much Google has been able to modify and reinforce its content-moderating guardrails.

And the risks extend beyond generating images of people as well. As mentioned in the fourth bullet point of the Key Takeaways section beginning this article, the A.I. can faithfully reproduce things like corporate logos, which may be a problem.

It may also generate copyrighted material that a user may not be familiar with. For instance, Sonic the Hedgehog.

Gems, Jewels, and Diamonds in the Rough

Companies with access to the Gems feature will have a lot to experiment with.

For instance, a corporation can create a Gem to help internal-memo writers with conforming to the company’s “house style”.

Some of the more interesting uses have to do with things like breaking down complex topics and helping with coding. Generally speaking, A.I. makes for a better coder than a writer, so if you have some programming tasks in your company, then A.I. can cut down on a lot of time.

The Final Key Takeaway

For business owners curious about Google’s ultimate standing in the A.I. race, it is the opinion of the writer of this blog that Google’s delayed start has been haunting it every step of the way.

Famously–to this writer, at least–, Google declared an internal “code red” on A.I. development after seeing just how advanced ChatGPT was after the debut. Surely, some of that worry had to do with Google’s competitor Microsoft being a massive investor in OpenAI, the developer of ChatGPT.

What Google has shown time and time again, whether it was with the Bard fiasco or the AI Overviews fiasco, is that it is simply lagging behind, and desperate to take some reputational hits for the sake of getting some wide public testing of its A.I. to gather important insights.