Imagen 2: a revolutionary text-to-image AI

Imagen 2 is a text-to-image scattering model developed by Google that features an unprecedented level of photorealism and deep language understanding. This model uses the power of large transformative language models to understand text and diffusion models to generate high-quality images.

Imagen improves image-text alignment

Our main finding is that large generic language models (such as T5), pre-trained on text-only corpora, are surprisingly effective at encoding text for image synthesis. However, by increasing the size of the language model, we found an improvement in sample fidelity and image-text alignment.

Google’s AI breaks records on COCO dataset

Imagen 2 achieves a new FID score of 7.27 on the COCO dataset, without ever training on COCO. Human raters also found that Imagen’s samples were equivalent to the COCO data itself in image-text alignment. We also used DrawBench, a comprehensive and challenging reference for text-image models. To compare Imagen to other recent methods and human reviewers preferred Google’s AI to other models in side-by-side comparisons.

Try Imagen 2 with Vertex AI

One of the best things about Imagen is that this artificial intelligence is already accessible via the Vertex AI platform. So not only can you quickly generate images, but you can also use its API easily and in a secure environment.

More technical information about imagen 2 here.