Imagen: Google's Image Generator Now Accessible to Everyone via Gemini

Google has made Imagen, its image generator via Gemini, available to the general public after I/O 2024. More photorealistic and integrated with SynthID to combat deepfakes, it still remains limited compared to Midjourney.

On October 9, Google opened its image generator, Imagen, to everyone via the Gemini platform, offering a new visual creation experience. As easily accessible as DALL-E is in ChatGPT, Imagen now allows any user to generate images in seconds, directly from the Gemini interface.

Unveiled in May during the I/O 2024 conference, this feature was initially reserved for a limited number of users. Today, it becomes available to a much wider audience. To generate an image, simply describe what you want in Gemini, and the model takes care of the rest, making visual creation simple and fast, even for novices.

Improvements in this new version

The new version of Imagen presents several notable improvements over its previous versions. First, the overall quality of the images has been significantly enhanced, with a more photorealistic rendering, reaching a level of realism comparable to tools like Midjourney or Flux, which are among the best on the market.

Additionally, Imagen now offers a greater diversity of artistic styles, allowing users to further customize their creations according to their desires. But the major innovation lies in the integration of SynthID technology, a system designed to invisibly mark generated images. This feature aims to combat disinformation and prevent the spread of deepfakes, by making it easy to identify artificially created content. With this advancement, Google enhances the security and transparency around images produced with Imagen.

Notable limitations

Despite its advancements, Imagen still has several limitations. For example, generating photorealistic people is only accessible to Gemini Advanced, Business, or Enterprise account holders. Moreover, even for these users, the model does not allow the creation of clearly identifiable individuals, representation of minors, or generation of bloody, violent, or sexual scenes, in order to adhere to strict ethical standards.

Some features remain reserved for premium subscribers, which limits the experience for standard users. Furthermore, Imagen currently only generates square images, and the option for inpainting — retouching generated images — is not yet available, although it is planned for a future update. A hyperlink directs users to a glossary explaining this technical term for those unfamiliar.

The test

During my test of Imagen, I noted a certain speed of execution. In just about ten seconds, the image is generated, which is quite satisfying. As for the quality of the images, it is interesting, although the photorealism does not yet rival that offered by tools like Midjourney.

The reproduction of people is not yet possible in the standard version, but Google announces that this feature will soon be available for Gemini Advanced users. Another limitation to note: Imagen only allows generating one image at a time, which can hinder productivity in some projects.

In terms of creation with artistic styles, the results are variable. Vector images are successful, but more specific styles like Chinese ink are not supported, while the psychedelic style yields interesting results. However, inserting a link as a style reference does not work.

Regarding the respect of prompts, Imagen tends to simplify descriptions. For example, a detailed scene describing a dog in a dense forest or a Yorkshire in a supermarket is simplified, thus reducing the richness of the visual.

Verdict

My verdict on Imagen is rather mixed. Interesting, yes, but mainly if you don't have other alternatives at hand. The tool is accessible and easy to use, making it a good option for casual or less demanding users. However, Imagen is clearly not designed for professional visual creators, who, unsurprisingly, will continue to prefer more advanced tools like Midjourney.

Indeed, the features remain very limited. For those seeking extensive control over their creations, Imagen does not meet expectations. Here are the main limitations:

• No format management: it's impossible to choose the size or dimensions of the image, remaining limited to square formats.
• No inpainting or outpainting for now, two essential tools for retouching or extending an image.
• No creation or persistence of characters, a major drawback for those wanting to create recurring or unique characters.
• Finally, there is no possibility to continue a style from a link, which limits customization and adaptation to existing works or styles.

In summary, Imagen is a good tool for quick and simple creation, but it lacks depth for advanced creators who will need more control and features.

By Brice Matter

____________________________

To stay updated on generative AI news, subscribe here to the Gennn Newsletter!