Photo: Google Getty Images
Google has officially released its latest AI image generator, Imagen 3, to the public in the United States. This powerful tool, now accessible through Google’s AI Test Kitchen service, allows users to create detailed and realistic images by simply typing a prompt and waiting about 30 seconds for the results.
First announced during Google’s I/O keynote in May 2024, Imagen 3 became widely available last week after Google published a research paper detailing its capabilities. According to Google, Imagen 3 is a state-of-the-art model that outperforms many of its competitors. “Imagen 3 is our highest quality text-to-image model, capable of generating images with even better detail, richer lighting, and fewer distracting artifacts than our previous models,” the company stated.
How Imagen 3 Works
Imagen 3 functions similarly to other AI image generators, such as OpenAI’s DALL-E and Midjourney. Users input a textual description, and the model generates an image that matches the prompt. PetaPixel, a technology news website, tested Imagen 3 and found it to be a strong competitor in the field, noting that it produces high-quality images for a wide variety of prompts.
The model’s ability to understand and respond to prompts has been significantly improved, allowing it to generate images with a wide range of visual styles and capture small details from more complex prompts. Google emphasizes that this makes Imagen 3 stand out from previous models.
Editing Capabilities and Restrictions
In addition to creating images, Imagen 3 offers editing options through a technique known as “inpainting.” This feature allows users to select a part of an image and modify it by typing in a desired change, giving them more control over the final output.
However, Google has placed certain restrictions on what users can generate with Imagen 3. For example, PetaPixel was unable to produce an image of “Kamala Harris and Donald Trump holding hands” or “A Californian landscape in the style of Ansel Adams.” These limitations are in place to prevent the generation of potentially sensitive or controversial content.
Despite these restrictions, some users have found workarounds. For instance, by requesting “a dramatic black and white photo taken in 1942 of the Grand Teton National Park in Wyoming,” they were able to create an image reminiscent of Ansel Adams’ iconic photography.
Similarly, The Verge, another technology news outlet, bypassed copyright restrictions by asking for “an image of a cartoonish blue hedgehog running in a field,” which resulted in an image similar to Sonic the Hedgehog.
Concerns Over Copyright and Bias
While Google has shared some information about how Imagen 3 was trained, there are still concerns regarding the data used. Google mentioned that the model was trained on a large dataset consisting of images, text, and associated annotations, but it is highly likely that the dataset includes copyrighted photos.
Earlier this year, Google faced criticism for its AI image generator on the Gemini platform, which was accused of overcorrecting for biases, leading to the erasure of white people from generated images. This controversy resulted in Google removing the image generator entirely. With Imagen 3, Google appears to be taking a more cautious approach, including restrictions and safeguards to prevent similar issues.
Free Access and Future Potential
One of the most appealing aspects of Imagen 3 is that it is currently free to use, unlike some of its competitors. This makes it accessible to a wider audience, allowing more people to explore its capabilities and contribute to the ongoing development of AI-generated content.
As AI technology continues to advance, tools like Imagen 3 represent the cutting edge of what is possible in digital creativity. By offering this powerful tool for free, Google is not only showcasing its technological prowess but also opening the door for users to experiment with and shape the future of AI-driven image creation.
If you want to try Imagen 3, you should head to the DeepMind website.