First look into imagen3: Google's latest text-to-image model

· 2 min read
VideoFX

Recently, the AI Labs account announced that early testers have gained access to Imagen3 through the ImageFX tool on AI Labs. Additionally, the VideoFX tool, which features a storyboard capability, has been updated. This storyboard feature allows users to create multiple images and then generate a video based on these images, telling a story. Significantly, it also relies on the Imagen3 model by default for image generation.

Previously, the storyboard feature was not functioning as intended. However, with the release of Imagen3 in the ImageFX tool, it appears that Imagen3 is now available in the storyboard as well. Reverse engineers have managed to generate images using the latest model. The model, code-named Juno, is confirmed to be different from Imagen2. Moreover, prompts that previously yielded weak results with Imagen2 show significant improvement with Imagen3, which produces much better images.

VideoFX

One major enhancement is the ability to generate detailed and high-quality images of humans, a challenge for previous models. However, text generation still needs improvement; for instance, Leonardo Phoenix performs better in this area.

Overall, take a look at these images and feel free to drop your prompts if you'd like us to test them with Imagen3.