Breakthrough in AI - Google's new image AI 'Imagen 4' finally spells correctly
The impressive accuracy of text generated by Imagen 4, unveiled at Google I/O 2025, marks a significant breakthrough.
 |
This comic, generated by Google's Imagen 4, demonstrates the AI's improved text rendering capabilities Image: Google |
For years, a persistent hurdle in AI-generated imagery has been the uncanny, often comical, misspellings and garbled text. We've all seen the memes – signs with nonsensical words or logos with a jumble of letters. This limitation has historically confined image AI to purely visual applications, sidestepping any need for legible text. Now, it seems Google has given its AI a truly remarkable spelling lesson, opening up a universe of new possibilities.
During the official Google I/O event on Monday, Google highlighted this monumental stride. The company introduced this new version of its image generation model, Imagen 4, which drastically improves text accuracy within images. This represents a huge improvement over other image generation models, especially the previous Imagen 3. In a recent blog post, Eli Collins, VP of Product at Google DeepMind, emphasized that Imagen 4 "combines speed with precision to create stunning images." He further elaborated on its "remarkable clarity in fine details like intricate fabrics, water droplets, and animal fur," noting its excellence in both photorealistic and abstract styles. Indeed, sample images released by Google are genuinely impressive, showcasing a whale gracefully breaching water and a chameleon with astonishing realism.
 |
A whale breaching in a city street, striking example of Imagen 4's photorealistic capabilities. Image: Google |
But the true showstopper, especially for those of us deeply immersed in content creation, is the promise of "significantly better at spelling and typography." Collins explicitly stated that this advancement makes it considerably easier to create practical items like greeting cards, posters, and comics – applications where accurate text isn't merely a bonus, but a fundamental requirement. Examples provided by Google showcase this promising aspect regarding spelling, such as a perfectly legible short comic and a tiny, readable font on a mock stamp. This is a noticeable leap forward, addressing a long-standing challenge in AI image generation.
 |
Image of a carton of eggs with a "Golden Yolk Farm Eggs" label generated by Imagen 4. Image: Google |
While the images presented instill great confidence that Google AI is gaining a significant edge over competitors in image generation, especially with these new features and enhanced spelling accuracy, the real test will undoubtedly come when the model is in the hands of regular users.
Availability and the Road Ahead for Imagen 4
Imagen 4 began rolling out on May 20th and will be integrated into popular Google platforms, including the Gemini app, Whisk, and Vertex AI. Furthermore, its powerful capabilities will extend across Workspace applications like Slides, Vids, and Docs, making this cutting-edge AI tool accessible to a broader audience. Google also hinted at a "fast variant" of Imagen 4 coming "soon," which is projected to be "up to 10x faster than Imagen 3," promising even more rapid prototyping and iteration.
