Google Unveils Imagen 3: The Ultimate Text-to-Image Model Now in Private Preview

Google is set to launch an impressive update to its text-to-image model, Imagen 3, which promises enhanced detail, improved natural language understanding, and superior text rendering. Currently available for select creators in a private preview on ImageFX, developers can join a waitlist for access. Additionally, Imagen 3 is anticipated to be integrated into Vertex AI.

“This is our most advanced image generation model to date,” said Douglas Eck, Senior Research Director of Google DeepMind, during the company’s I/O developer conference. “Imagen 3 offers more photorealism, richer details, and significantly fewer visual artifacts or distortions. It understands prompts in a conversational manner—the more creativity and detail you provide, the better the output. Imagen 3 excels at incorporating subtle elements in longer prompts and represents our best effort yet in rendering text, a common challenge for image generation models.”

The announcement comes just six months after Google made Imagen 2 widely available on Vertex AI, following its earlier rollout of text-to-live capabilities in April. Staying ahead in the competitive landscape of AI image generation tools—such as OpenAI's DALL-E, Midjourney, Adobe's Firefly, Meta's AI, and Microsoft's Designer—is crucial for Google.

However, Google's journey in this domain hasn't been without challenges. Earlier this year, the company faced backlash for its AI-generated images, which some critics labeled as “too woke.” This included depictions of the U.S. Founding Fathers with a Black man and World War II German soldiers alongside a Black man and Asian woman. Google acknowledged the misstep, with CEO Sundar Pichai reaffirming the apology during an interview with Bloomberg.

The reveal of Imagen 3 is one of several exciting updates from Google I/O, which also includes the launch of a new video generation model, Veo.