Google opens Imagen 3 access to all US users

Google opened Imagen 3, its text-to-image model, to all US users through ImageFX, roughly three months after first demonstrating it at Google I/O in May 2024. Access had previously been limited to a waitlist.

Google described Imagen 3 as its highest-quality image generator to date, with particular improvement in two areas that had lagged competitors: rendering legible text within generated images, and reducing the visual artefacts — warped hands, garbled detail, repeating patterns — that had dogged earlier diffusion models, including Google’s own original Imagen. The company said the model was preferred over other state-of-the-art image generators in its own evaluations, without publishing the comparison in detail. Imagen 3 retained the safety-first posture Google had taken with its media models generally, built-in restrictions and prompt denial for a range of sensitive categories, in contrast to the comparatively permissive approach of rivals such as Midjourney.

The staged rollout — announcement at a developer conference, followed by a limited then general release — was typical of how Google shipped generative-media models through 2024, a more cautious cadence than the near-simultaneous announce-and-ship pattern OpenAI and Midjourney used for image tools over the same period.