The best AI image generators have been in the news recently, but not always for the right reasons – with Elon Musk’s Grok-2 AI producing all sorts of controversy over its lack of guardrails. Google probably wants to shift the conversation in a more positive direction, as the company has just expanded access to its Imagen 3 generator to everyone in the US, massively ramping up who can use the AI tool.
Google’s accompanying research paper describes Imagen 3 as “a latent diffusion model that generates high quality images from text prompts,” with the paper adding that “Imagen 3 is preferred over other state-of-the-art models at the time of evaluation.” It can be used to whip up images based on text entered by the user, with some restrictions on offensive or illegal content, making it a rival to popular alternatives like Midjourney.
The tool was originally launched in May 2024 at the Google I/O event, but at that time it was limited to chosen users of Google’s Vertex AI. Now, anyone in the U.S. can try it out if they like, suggesting that Google is much more confident in its tool’s ability to create reliable images and avoid some of the problems that have plagued AI image generators for so long.
So far, it seems that the reaction to Imagen 3 has been mixed. Several users have noticed that it seems to be much more sensitive to user prompts than before, censoring words that might not have been blocked in Imagen 2.
For instance, one user on Reddit said: “Really have to put in extra work to achieve what I used to get and a random word like “sock” or “water” will trigger the censorship filter which is far more sensitive to benign words.” For another, Imagen 3 refused to draw innocent-sounding ideas such as a cyborg or a man crossing his arms. There are positives, though, with one of the aforementioned users noting its high-quality output and “amazing texture and word recognition.”
How to try Imagen 3
If you want to take Imagen 3 for a spin, you’ll first need to be based in the U.S. to gain access. If you are, you can head to Google’s AI Test Kitchen website and sign in with your Google account. Once that’s done, you should be able to start using the new image-generation model.
User reports that Imagen 3 is more restrictive than its predecessor are a useful illustration of the current state of AI image generators. A huge amount of controversy has been created by these tools in recent months, and it sometimes seems that not a week goes by without some AI-generated image stirring up the internet’s emotions.
The latest culprit has been Elon Musk’s Grok-2 AI. The tool – which is limited to premium users of X (formerly Twitter), was just updated to its second version and generates images using the open-source Flux. Users found it had very few limits on what could be created, leading to people drawing up pictures of Donald Trump and Kamala Harris piloting a plane on 9/11, or clearly copyrighted images being generated with seemingly no push back from Grok.
With its restrictive perspective on image generation, it seems that Google has set Imagen 3 to err on the side of caution for now. Whether that will change as the model is tweaked and refined remains to be seen, but for with far more people now able to access it, perhaps we won’t have to wait long to see how far its boundaries can be pushed.
+ There are no comments
Add yours