![images 3.fill .size 2000x1108.v1723834570 images 3.fill .size 2000x1108.v1723834570](https://utterbuzz.com/wp-content/uploads/2024/08/images-3.fill_.size_2000x1108.v1723834570-640x355.png)
Google’s latest AI image generation model, Imagen 3, is now publicly available in the United States, and all you need to try it out is a free Google account. As spotted by VentureBeat, the company quietly opened up its model this week, publishing its research in a paper on Tuesday. It comes two months after the company first announced the new model at Google I/O back in May.
In a post on Hugging Face, a machine learning platform, Google researchers said, “We introduce Imagen 3, a latent diffusion model that generates high quality images from text prompts. We describe our quality and responsibility evaluations. Imagen 3 is preferred over other state-of-the-art (SOTA) models at the time of evaluation. In addition, we discuss issues around safety and representation, as well as methods we used to minimize the potential harm of our models.”
From some initial amateurish tests on my end, the model seems pretty solid. When asking for photorealistic images, the results are relatively high quality, with some decent stabs at realism that could fool people on first glance. (I was particularly impressed with the quality of images when I prompted the model to create a 35mm film look.) Imagen 3 also highlights particular parts of the prompt that affected the output, so you can adjust those if you don’t like the way the image came out.
![prompt for a 35mm look for a soccer game](https://lifehacker.com/imagery/articles/01J5E5TP9JMCE3SG8GRA6H4PE6/images-3.fill.size_2000x1108.v1723834570.png)
Credit: Jake Peterson/Google
However, Imagen 3’s offerings still display the telltale signs of AI-generated images. In some photos, hands have too many fingers, faces are distorted, and text doesn’t make sense. (Although the model was able to reproduce the “Coca-Cola” and “Canon” logos with trademark-infringing accuracy.)
![ai image of a coke bottle](https://lifehacker.com/imagery/articles/01J5E5TP9JMCE3SG8GRA6H4PE6/images-2.fill.size_2000x1165.v1723834570.png)
Credit: Jake Peterson/Google
Google isn’t the only tech company to release a new image model this week. X recently pushed a new beta for Grok, the company’s AI chatbot, and with it, an image generator built with seemingly very few restrictions. Users (specifically, Lifehacker’s Michelle Ehrhardt) have used Grok to generate everything from Taylor Swift in a MAGA hat to Pikachu holding an AK-47.
Imagen 3, on the other hand, displays obvious guardrails. When I attempt to prompt Google’s image generator with anything controversial, it stops itself and politely directs me to Imagen 3’s FAQ to understand why my prompt wasn’t appropriate. It also refuses to generate copyrighted content, but can be tricked into producing it with the right prompts. As noted above, I was able to generate logos, and was even able to get it to spit out trademarked characters like Mario and Pikachu, even if I couldn’t get them to engage in a gunfight.
![AI generated Pikachu](https://lifehacker.com/imagery/articles/01J5E5TP9JMCE3SG8GRA6H4PE6/images-5.fill.size_2000x1165.v1723834570.png)
Credit: Jake Peterson/Google
How to try Imagen 3
Google’s new AI image generator is free to try for any U.S. user with a Google Account. To do so, head to ImageFX in Google’s AI Test Kitchen. Sign in with your Google Account, then get prompting.