Google DeepMind recently published the paper for it’s new state-of-the-art AI image generation model,

Google DeepMind recently published the paper for it’s new state-of-the-art AI image generation model, Imagen 3, flexing that it beat DALL-E 3, Midjourney v6, and Stable Diffusion 3 in human performance evaluations.

The details:

The human evaluations asked participants to rank their preferred models for overall quality and adherence to detailed prompts.
Imagen 3 excelled particularly in generating high-quality, realistic images that closely match long and complex text descriptions.
Despite its capability to accurately generate photorealistic images, it struggles with certain tasks requiring numerical reasoning, understanding scale, and depicting actions.
Imagen 3 is available to try via ImageFX and Vertex AI.

Why it matters: Google struggled to find its footing early in the AI text-to-image category, but with its latest Imagen 3 release, it’s beating the top tools in the space. It’s another win for Google after also beating OpenAI in the race to widespread rollouts of advanced voice AI just yesterday.

6 comments